avoid sprintf and strcpy with flex arrays
[git/git.git] / sha1_file.c
CommitLineData
0fcfd160
LT
1/*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 *
6 * This handles basic git sha1 object files - packing, unpacking,
7 * creation etc.
8 */
0fcfd160 9#include "cache.h"
6eac50d8 10#include "string-list.h"
697cc8ef 11#include "lockfile.h"
1f688557 12#include "delta.h"
a733cb60 13#include "pack.h"
8e440259
PE
14#include "blob.h"
15#include "commit.h"
4dd1fbc7 16#include "run-command.h"
8e440259
PE
17#include "tag.h"
18#include "tree.h"
c879daa2 19#include "tree-walk.h"
f35a6d3b 20#include "refs.h"
70f5d5d3 21#include "pack-revindex.h"
628522ec 22#include "sha1-lookup.h"
568508e7 23#include "bulk-checkin.h"
090ea126 24#include "streaming.h"
543c5caa 25#include "dir.h"
0fcfd160 26
144bde78
LT
27#ifndef O_NOATIME
28#if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
29#define O_NOATIME 01000000
30#else
31#define O_NOATIME 0
32#endif
33#endif
34
28bd70d8
JN
35#define SZ_FMT PRIuMAX
36static inline uintmax_t sz_fmt(size_t s) { return s; }
e05db0fd 37
96f1e58f 38const unsigned char null_sha1[20];
88cd621d 39
c597ba80
NTND
40/*
41 * This is meant to hold a *small* number of objects that you would
42 * want read_sha1_file() to be able to return, but yet you do not want
43 * to write them into the object store (e.g. a browse-only
44 * application).
45 */
46static struct cached_object {
47 unsigned char sha1[20];
48 enum object_type type;
49 void *buf;
50 unsigned long size;
51} *cached_objects;
52static int cached_object_nr, cached_object_alloc;
53
54static struct cached_object empty_tree = {
dab0d410 55 EMPTY_TREE_SHA1_BIN_LITERAL,
c597ba80
NTND
56 OBJ_TREE,
57 "",
58 0
59};
60
1b1005d1
MH
61/*
62 * A pointer to the last packed_git in which an object was found.
63 * When an object is sought, we look in this packfile first, because
64 * objects that are looked up at similar times are often in the same
65 * packfile as one another.
66 */
c01f51cc
NTND
67static struct packed_git *last_found_pack;
68
c597ba80
NTND
69static struct cached_object *find_cached_object(const unsigned char *sha1)
70{
71 int i;
72 struct cached_object *co = cached_objects;
73
74 for (i = 0; i < cached_object_nr; i++, co++) {
75 if (!hashcmp(co->sha1, sha1))
76 return co;
77 }
78 if (!hashcmp(sha1, empty_tree.sha1))
79 return &empty_tree;
80 return NULL;
81}
82
90a6464b
JH
83int mkdir_in_gitdir(const char *path)
84{
85 if (mkdir(path, 0777)) {
86 int saved_errno = errno;
87 struct stat st;
88 struct strbuf sb = STRBUF_INIT;
89
90 if (errno != EEXIST)
91 return -1;
92 /*
93 * Are we looking at a path in a symlinked worktree
94 * whose original repository does not yet have it?
95 * e.g. .git/rr-cache pointing at its original
96 * repository in which the user hasn't performed any
97 * conflict resolution yet?
98 */
99 if (lstat(path, &st) || !S_ISLNK(st.st_mode) ||
100 strbuf_readlink(&sb, path, st.st_size) ||
101 !is_absolute_path(sb.buf) ||
102 mkdir(sb.buf, 0777)) {
103 strbuf_release(&sb);
104 errno = saved_errno;
105 return -1;
106 }
107 strbuf_release(&sb);
108 }
109 return adjust_shared_perm(path);
110}
111
0be0521b 112enum scld_error safe_create_leading_directories(char *path)
b2cb9425 113{
26c8ae2a 114 char *next_component = path + offset_1st_component(path);
0be0521b 115 enum scld_error ret = SCLD_OK;
67d42212 116
0be0521b 117 while (ret == SCLD_OK && next_component) {
f0502332 118 struct stat st;
0f527403 119 char *slash = next_component, slash_character;
f0502332 120
0f527403
MH
121 while (*slash && !is_dir_sep(*slash))
122 slash++;
123
124 if (!*slash)
b2cb9425 125 break;
bf10cf70 126
26c8ae2a 127 next_component = slash + 1;
0f527403 128 while (is_dir_sep(*next_component))
bf10cf70 129 next_component++;
26c8ae2a 130 if (!*next_component)
5f0bdf50 131 break;
831651fd 132
0f527403 133 slash_character = *slash;
831651fd 134 *slash = '\0';
67d42212
JR
135 if (!stat(path, &st)) {
136 /* path exists */
9e6f885d 137 if (!S_ISDIR(st.st_mode))
0be0521b 138 ret = SCLD_EXISTS;
53a39721 139 } else if (mkdir(path, 0777)) {
928734d9 140 if (errno == EEXIST &&
9e6f885d 141 !stat(path, &st) && S_ISDIR(st.st_mode))
928734d9 142 ; /* somebody created it since we checked */
18d37e86
MH
143 else if (errno == ENOENT)
144 /*
145 * Either mkdir() failed because
146 * somebody just pruned the containing
147 * directory, or stat() failed because
148 * the file that was in our way was
149 * just removed. Either way, inform
150 * the caller that it might be worth
151 * trying again:
152 */
153 ret = SCLD_VANISHED;
9e6f885d 154 else
0be0521b 155 ret = SCLD_FAILED;
53a39721 156 } else if (adjust_shared_perm(path)) {
0be0521b 157 ret = SCLD_PERMS;
457f06d6 158 }
0f527403 159 *slash = slash_character;
b2cb9425 160 }
9e6f885d 161 return ret;
b2cb9425 162}
723c31fe 163
0be0521b 164enum scld_error safe_create_leading_directories_const(const char *path)
8e21d63b
JK
165{
166 /* path points to cache entries, so xstrdup before messing with it */
167 char *buf = xstrdup(path);
0be0521b 168 enum scld_error result = safe_create_leading_directories(buf);
8e21d63b
JK
169 free(buf);
170 return result;
171}
172
ace1534d
JH
173static void fill_sha1_path(char *pathbuf, const unsigned char *sha1)
174{
175 int i;
176 for (i = 0; i < 20; i++) {
177 static char hex[] = "0123456789abcdef";
178 unsigned int val = sha1[i];
179 char *pos = pathbuf + i*2 + (i > 0);
180 *pos++ = hex[val >> 4];
181 *pos = hex[val & 0xf];
182 }
183}
184
30d6c6ea 185const char *sha1_file_name(const unsigned char *sha1)
0fcfd160 186{
560fb6a1
JK
187 static char buf[PATH_MAX];
188 const char *objdir;
189 int len;
0fcfd160 190
560fb6a1
JK
191 objdir = get_object_directory();
192 len = strlen(objdir);
193
194 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
195 if (len + 43 > PATH_MAX)
196 die("insanely long object directory %s", objdir);
197 memcpy(buf, objdir, len);
198 buf[len] = '/';
199 buf[len+3] = '/';
200 buf[len+42] = '\0';
201 fill_sha1_path(buf + len + 1, sha1);
202 return buf;
0fcfd160
LT
203}
204
d40d535b
MH
205/*
206 * Return the name of the pack or index file with the specified sha1
207 * in its filename. *base and *name are scratch space that must be
208 * provided by the caller. which should be "pack" or "idx".
209 */
633f43e1 210static char *sha1_get_pack_name(const unsigned char *sha1,
ac5190cc
JK
211 struct strbuf *buf,
212 const char *which)
bf592c50 213{
ac5190cc
JK
214 strbuf_reset(buf);
215 strbuf_addf(buf, "%s/pack/pack-%s.%s", get_object_directory(),
216 sha1_to_hex(sha1), which);
217 return buf->buf;
bf592c50
DB
218}
219
633f43e1 220char *sha1_pack_name(const unsigned char *sha1)
bf592c50 221{
ac5190cc
JK
222 static struct strbuf buf = STRBUF_INIT;
223 return sha1_get_pack_name(sha1, &buf, "pack");
633f43e1 224}
bf592c50 225
633f43e1
HO
226char *sha1_pack_index_name(const unsigned char *sha1)
227{
ac5190cc
JK
228 static struct strbuf buf = STRBUF_INIT;
229 return sha1_get_pack_name(sha1, &buf, "idx");
bf592c50
DB
230}
231
d5a63b99
JH
232struct alternate_object_database *alt_odb_list;
233static struct alternate_object_database **alt_odb_tail;
ace1534d 234
ddd5d056
JH
235/*
236 * Prepare alternate object database registry.
d5a63b99
JH
237 *
238 * The variable alt_odb_list points at the list of struct
239 * alternate_object_database. The elements on this list come from
240 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
241 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
1494e038
JH
242 * whose contents is similar to that environment variable but can be
243 * LF separated. Its base points at a statically allocated buffer that
d5a63b99
JH
244 * contains "/the/directory/corresponding/to/.git/objects/...", while
245 * its name points just after the slash at the end of ".git/objects/"
246 * in the example above, and has enough space to hold 40-byte hex
247 * SHA1, an extra slash for the first level indirection, and the
248 * terminating NUL.
ddd5d056 249 */
539e7506
EK
250static int link_alt_odb_entry(const char *entry, const char *relative_base,
251 int depth, const char *normalized_objdir)
ace1534d 252{
c2f493a4
MW
253 struct alternate_object_database *ent;
254 struct alternate_object_database *alt;
5bdf0a84
HW
255 int pfxlen, entlen;
256 struct strbuf pathbuf = STRBUF_INIT;
d5a63b99 257
85dadc38 258 if (!is_absolute_path(entry) && relative_base) {
5bdf0a84
HW
259 strbuf_addstr(&pathbuf, real_path(relative_base));
260 strbuf_addch(&pathbuf, '/');
c2f493a4 261 }
6eac50d8 262 strbuf_addstr(&pathbuf, entry);
c2f493a4 263
5bdf0a84
HW
264 normalize_path_copy(pathbuf.buf, pathbuf.buf);
265
266 pfxlen = strlen(pathbuf.buf);
267
268 /*
269 * The trailing slash after the directory name is given by
270 * this function at the end. Remove duplicates.
271 */
272 while (pfxlen && pathbuf.buf[pfxlen-1] == '/')
273 pfxlen -= 1;
274
275 entlen = pfxlen + 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
276 ent = xmalloc(sizeof(*ent) + entlen);
277 memcpy(ent->base, pathbuf.buf, pfxlen);
278 strbuf_release(&pathbuf);
c2f493a4
MW
279
280 ent->name = ent->base + pfxlen + 1;
281 ent->base[pfxlen + 3] = '/';
282 ent->base[pfxlen] = ent->base[entlen-1] = 0;
283
284 /* Detect cases where alternate disappeared */
90b4a71c 285 if (!is_directory(ent->base)) {
c2f493a4
MW
286 error("object directory %s does not exist; "
287 "check .git/objects/info/alternates.",
288 ent->base);
289 free(ent);
290 return -1;
291 }
292
293 /* Prevent the common mistake of listing the same
294 * thing twice, or object directory itself.
295 */
296 for (alt = alt_odb_list; alt; alt = alt->next) {
80b47854
RS
297 if (pfxlen == alt->name - alt->base - 1 &&
298 !memcmp(ent->base, alt->base, pfxlen)) {
c2f493a4
MW
299 free(ent);
300 return -1;
301 }
302 }
539e7506 303 if (!strcmp_icase(ent->base, normalized_objdir)) {
c2f493a4
MW
304 free(ent);
305 return -1;
306 }
307
308 /* add the alternate entry */
309 *alt_odb_tail = ent;
310 alt_odb_tail = &(ent->next);
311 ent->next = NULL;
312
313 /* recursively add alternates */
314 read_info_alternates(ent->base, depth + 1);
315
316 ent->base[pfxlen] = '/';
317
318 return 0;
319}
320
c5950164 321static void link_alt_odb_entries(const char *alt, int len, int sep,
c2f493a4
MW
322 const char *relative_base, int depth)
323{
6eac50d8
MH
324 struct string_list entries = STRING_LIST_INIT_NODUP;
325 char *alt_copy;
326 int i;
539e7506 327 struct strbuf objdirbuf = STRBUF_INIT;
c2f493a4
MW
328
329 if (depth > 5) {
330 error("%s: ignoring alternate object stores, nesting too deep.",
331 relative_base);
332 return;
333 }
334
f655651e 335 strbuf_add_absolute_path(&objdirbuf, get_object_directory());
539e7506
EK
336 normalize_path_copy(objdirbuf.buf, objdirbuf.buf);
337
c5950164 338 alt_copy = xmemdupz(alt, len);
6eac50d8
MH
339 string_list_split_in_place(&entries, alt_copy, sep, -1);
340 for (i = 0; i < entries.nr; i++) {
341 const char *entry = entries.items[i].string;
342 if (entry[0] == '\0' || entry[0] == '#')
9577e7e3 343 continue;
6eac50d8
MH
344 if (!is_absolute_path(entry) && depth) {
345 error("%s: ignoring relative alternate object store %s",
346 relative_base, entry);
347 } else {
539e7506 348 link_alt_odb_entry(entry, relative_base, depth, objdirbuf.buf);
9577e7e3 349 }
9577e7e3 350 }
6eac50d8
MH
351 string_list_clear(&entries, 0);
352 free(alt_copy);
539e7506 353 strbuf_release(&objdirbuf);
d5a63b99
JH
354}
355
5e73633d 356void read_info_alternates(const char * relative_base, int depth)
d5a63b99 357{
9577e7e3 358 char *map;
dc49cd76 359 size_t mapsz;
d5a63b99 360 struct stat st;
5015f01c 361 char *path;
c2f493a4 362 int fd;
d5a63b99 363
5015f01c 364 path = xstrfmt("%s/info/alternates", relative_base);
f4e51683 365 fd = git_open_noatime(path);
5015f01c 366 free(path);
d5a63b99
JH
367 if (fd < 0)
368 return;
369 if (fstat(fd, &st) || (st.st_size == 0)) {
370 close(fd);
9a217f2a 371 return;
ace1534d 372 }
dc49cd76
SP
373 mapsz = xsize_t(st.st_size);
374 map = xmmap(NULL, mapsz, PROT_READ, MAP_PRIVATE, fd, 0);
d5a63b99 375 close(fd);
d5a63b99 376
c5950164 377 link_alt_odb_entries(map, mapsz, '\n', relative_base, depth);
c2f493a4 378
dc49cd76 379 munmap(map, mapsz);
ace1534d
JH
380}
381
bef70b22
DB
382void add_to_alternates_file(const char *reference)
383{
384 struct lock_file *lock = xcalloc(1, sizeof(struct lock_file));
77b9b1d1
JK
385 char *alts = git_pathdup("objects/info/alternates");
386 FILE *in, *out;
387
388 hold_lock_file_for_update(lock, alts, LOCK_DIE_ON_ERROR);
389 out = fdopen_lock_file(lock, "w");
390 if (!out)
391 die_errno("unable to fdopen alternates lockfile");
392
393 in = fopen(alts, "r");
394 if (in) {
395 struct strbuf line = STRBUF_INIT;
396 int found = 0;
397
398 while (strbuf_getline(&line, in, '\n') != EOF) {
399 if (!strcmp(reference, line.buf)) {
400 found = 1;
401 break;
402 }
403 fprintf_or_die(out, "%s\n", line.buf);
404 }
405
406 strbuf_release(&line);
407 fclose(in);
408
409 if (found) {
410 rollback_lock_file(lock);
411 lock = NULL;
412 }
413 }
414 else if (errno != ENOENT)
415 die_errno("unable to read alternates file");
416
417 if (lock) {
418 fprintf_or_die(out, "%s\n", reference);
419 if (commit_lock_file(lock))
420 die_errno("unable to move new alternates file into place");
421 if (alt_odb_tail)
422 link_alt_odb_entries(reference, strlen(reference), '\n', NULL, 0);
423 }
424 free(alts);
bef70b22
DB
425}
426
fe1b2268 427int foreach_alt_odb(alt_odb_fn fn, void *cb)
d79796bc
JH
428{
429 struct alternate_object_database *ent;
fe1b2268 430 int r = 0;
d79796bc
JH
431
432 prepare_alt_odb();
fe1b2268
JK
433 for (ent = alt_odb_list; ent; ent = ent->next) {
434 r = fn(ent, cb);
435 if (r)
436 break;
437 }
438 return r;
d79796bc
JH
439}
440
c2f493a4
MW
441void prepare_alt_odb(void)
442{
554fe20d 443 const char *alt;
c2f493a4 444
7dc24aa5
SP
445 if (alt_odb_tail)
446 return;
447
c2f493a4
MW
448 alt = getenv(ALTERNATE_DB_ENVIRONMENT);
449 if (!alt) alt = "";
450
c2f493a4 451 alt_odb_tail = &alt_odb_list;
c5950164 452 link_alt_odb_entries(alt, strlen(alt), PATH_SEP, NULL, 0);
c2f493a4
MW
453
454 read_info_alternates(get_object_directory(), 0);
455}
456
3096b2ec 457/* Returns 1 if we have successfully freshened the file, 0 otherwise. */
33d4221c 458static int freshen_file(const char *fn)
ace1534d 459{
33d4221c
JK
460 struct utimbuf t;
461 t.actime = t.modtime = time(NULL);
462 return !utime(fn, &t);
0f4dc14a 463}
ace1534d 464
3096b2ec
JK
465/*
466 * All of the check_and_freshen functions return 1 if the file exists and was
467 * freshened (if freshening was requested), 0 otherwise. If they return
468 * 0, you should not assume that it is safe to skip a write of the object (it
469 * either does not exist on disk, or has a stale mtime and may be subject to
470 * pruning).
471 */
33d4221c
JK
472static int check_and_freshen_file(const char *fn, int freshen)
473{
474 if (access(fn, F_OK))
475 return 0;
3096b2ec 476 if (freshen && !freshen_file(fn))
33d4221c
JK
477 return 0;
478 return 1;
479}
480
481static int check_and_freshen_local(const unsigned char *sha1, int freshen)
482{
483 return check_and_freshen_file(sha1_file_name(sha1), freshen);
484}
485
486static int check_and_freshen_nonlocal(const unsigned char *sha1, int freshen)
0f4dc14a
BC
487{
488 struct alternate_object_database *alt;
9a217f2a 489 prepare_alt_odb();
d5a63b99 490 for (alt = alt_odb_list; alt; alt = alt->next) {
0f4dc14a 491 fill_sha1_path(alt->name, sha1);
33d4221c 492 if (check_and_freshen_file(alt->base, freshen))
c529d75a 493 return 1;
ace1534d 494 }
c529d75a 495 return 0;
ace1534d
JH
496}
497
33d4221c
JK
498static int check_and_freshen(const unsigned char *sha1, int freshen)
499{
500 return check_and_freshen_local(sha1, freshen) ||
501 check_and_freshen_nonlocal(sha1, freshen);
502}
503
504int has_loose_object_nonlocal(const unsigned char *sha1)
505{
506 return check_and_freshen_nonlocal(sha1, 0);
507}
508
0f4dc14a
BC
509static int has_loose_object(const unsigned char *sha1)
510{
33d4221c 511 return check_and_freshen(sha1, 0);
0f4dc14a
BC
512}
513
60bb8b14 514static unsigned int pack_used_ctr;
a53128b6
SP
515static unsigned int pack_mmap_calls;
516static unsigned int peak_pack_open_windows;
517static unsigned int pack_open_windows;
c7934306
SP
518static unsigned int pack_open_fds;
519static unsigned int pack_max_fds;
a53128b6 520static size_t peak_pack_mapped;
60bb8b14 521static size_t pack_mapped;
9a217f2a 522struct packed_git *packed_git;
1f688557 523
b79d18c9 524void pack_report(void)
a53128b6
SP
525{
526 fprintf(stderr,
e05db0fd
PR
527 "pack_report: getpagesize() = %10" SZ_FMT "\n"
528 "pack_report: core.packedGitWindowSize = %10" SZ_FMT "\n"
529 "pack_report: core.packedGitLimit = %10" SZ_FMT "\n",
9e42d6a1
SP
530 sz_fmt(getpagesize()),
531 sz_fmt(packed_git_window_size),
532 sz_fmt(packed_git_limit));
a53128b6
SP
533 fprintf(stderr,
534 "pack_report: pack_used_ctr = %10u\n"
535 "pack_report: pack_mmap_calls = %10u\n"
536 "pack_report: pack_open_windows = %10u / %10u\n"
e05db0fd
PR
537 "pack_report: pack_mapped = "
538 "%10" SZ_FMT " / %10" SZ_FMT "\n",
a53128b6
SP
539 pack_used_ctr,
540 pack_mmap_calls,
541 pack_open_windows, peak_pack_open_windows,
9e42d6a1 542 sz_fmt(pack_mapped), sz_fmt(peak_pack_mapped));
a53128b6
SP
543}
544
d40d535b
MH
545/*
546 * Open and mmap the index file at path, perform a couple of
547 * consistency checks, then record its information to p. Return 0 on
548 * success.
549 */
550static int check_packed_git_idx(const char *path, struct packed_git *p)
1f688557
JH
551{
552 void *idx_map;
42873078 553 struct pack_idx_header *hdr;
dc49cd76 554 size_t idx_size;
74e34e1f 555 uint32_t version, nr, i, *index;
f4e51683 556 int fd = git_open_noatime(path);
1f688557 557 struct stat st;
42873078 558
1f688557
JH
559 if (fd < 0)
560 return -1;
561 if (fstat(fd, &st)) {
562 close(fd);
563 return -1;
564 }
dc49cd76 565 idx_size = xsize_t(st.st_size);
2d88451b
SP
566 if (idx_size < 4 * 256 + 20 + 20) {
567 close(fd);
568 return error("index file %s is too small", path);
569 }
c4712e45 570 idx_map = xmmap(NULL, idx_size, PROT_READ, MAP_PRIVATE, fd, 0);
1f688557 571 close(fd);
1f688557 572
42873078
NP
573 hdr = idx_map;
574 if (hdr->idx_signature == htonl(PACK_IDX_SIGNATURE)) {
74e34e1f
NP
575 version = ntohl(hdr->idx_version);
576 if (version < 2 || version > 2) {
577 munmap(idx_map, idx_size);
6e1c2344 578 return error("index file %s is version %"PRIu32
74e34e1f
NP
579 " and is not supported by this binary"
580 " (try upgrading GIT to a newer version)",
581 path, version);
582 }
583 } else
584 version = 1;
df1b059d 585
1f688557 586 nr = 0;
42873078 587 index = idx_map;
74e34e1f
NP
588 if (version > 1)
589 index += 2; /* skip index header */
1f688557 590 for (i = 0; i < 256; i++) {
326bf396 591 uint32_t n = ntohl(index[i]);
2d88451b
SP
592 if (n < nr) {
593 munmap(idx_map, idx_size);
df1b059d 594 return error("non-monotonic index %s", path);
2d88451b 595 }
1f688557
JH
596 nr = n;
597 }
598
74e34e1f
NP
599 if (version == 1) {
600 /*
601 * Total size:
602 * - 256 index entries 4 bytes each
603 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
604 * - 20-byte SHA1 of the packfile
605 * - 20-byte SHA1 file checksum
606 */
607 if (idx_size != 4*256 + nr * 24 + 20 + 20) {
608 munmap(idx_map, idx_size);
eef427a0 609 return error("wrong index v1 file size in %s", path);
74e34e1f
NP
610 }
611 } else if (version == 2) {
612 /*
613 * Minimum size:
614 * - 8 bytes of header
615 * - 256 index entries 4 bytes each
616 * - 20-byte sha1 entry * nr
617 * - 4-byte crc entry * nr
618 * - 4-byte offset entry * nr
619 * - 20-byte SHA1 of the packfile
620 * - 20-byte SHA1 file checksum
621 * And after the 4-byte offset table might be a
622 * variable sized table containing 8-byte entries
623 * for offsets larger than 2^31.
624 */
625 unsigned long min_size = 8 + 4*256 + nr*(20 + 4 + 4) + 20 + 20;
1164f1e4
LT
626 unsigned long max_size = min_size;
627 if (nr)
628 max_size += (nr - 1)*8;
629 if (idx_size < min_size || idx_size > max_size) {
74e34e1f 630 munmap(idx_map, idx_size);
eef427a0 631 return error("wrong index v2 file size in %s", path);
74e34e1f 632 }
7109c889
JH
633 if (idx_size != min_size &&
634 /*
635 * make sure we can deal with large pack offsets.
636 * 31-bit signed offset won't be enough, neither
637 * 32-bit unsigned one will be.
638 */
639 (sizeof(off_t) <= 4)) {
640 munmap(idx_map, idx_size);
641 return error("pack too large for current definition of off_t in %s", path);
74e34e1f 642 }
2d88451b 643 }
1f688557 644
74e34e1f 645 p->index_version = version;
42873078
NP
646 p->index_data = idx_map;
647 p->index_size = idx_size;
57059091 648 p->num_objects = nr;
1f688557
JH
649 return 0;
650}
651
bc8e478a 652int open_pack_index(struct packed_git *p)
d079837e
SP
653{
654 char *idx_name;
9ae97018 655 size_t len;
d079837e
SP
656 int ret;
657
658 if (p->index_data)
659 return 0;
660
9ae97018
JK
661 if (!strip_suffix(p->pack_name, ".pack", &len))
662 die("BUG: pack_name does not end in .pack");
663 idx_name = xstrfmt("%.*s.idx", (int)len, p->pack_name);
d079837e
SP
664 ret = check_packed_git_idx(idx_name, p);
665 free(idx_name);
666 return ret;
667}
668
11daf39b
SP
669static void scan_windows(struct packed_git *p,
670 struct packed_git **lru_p,
671 struct pack_window **lru_w,
672 struct pack_window **lru_l)
1f688557 673{
11daf39b
SP
674 struct pack_window *w, *w_l;
675
676 for (w_l = NULL, w = p->windows; w; w = w->next) {
677 if (!w->inuse_cnt) {
678 if (!*lru_w || w->last_used < (*lru_w)->last_used) {
679 *lru_p = p;
680 *lru_w = w;
681 *lru_l = w_l;
54044bf8 682 }
54044bf8 683 }
11daf39b 684 w_l = w;
f9253394 685 }
11daf39b
SP
686}
687
7c3ecb32 688static int unuse_one_window(struct packed_git *current)
11daf39b
SP
689{
690 struct packed_git *p, *lru_p = NULL;
691 struct pack_window *lru_w = NULL, *lru_l = NULL;
692
693 if (current)
694 scan_windows(current, &lru_p, &lru_w, &lru_l);
695 for (p = packed_git; p; p = p->next)
696 scan_windows(p, &lru_p, &lru_w, &lru_l);
54044bf8
SP
697 if (lru_p) {
698 munmap(lru_w->base, lru_w->len);
699 pack_mapped -= lru_w->len;
700 if (lru_l)
701 lru_l->next = lru_w->next;
7c3ecb32 702 else
54044bf8 703 lru_p->windows = lru_w->next;
54044bf8 704 free(lru_w);
a53128b6 705 pack_open_windows--;
54044bf8
SP
706 return 1;
707 }
708 return 0;
f9253394
JH
709}
710
7c3ecb32 711void release_pack_memory(size_t need)
97bfeb34
SP
712{
713 size_t cur = pack_mapped;
7c3ecb32 714 while (need >= (cur - pack_mapped) && unuse_one_window(NULL))
97bfeb34
SP
715 ; /* nothing */
716}
717
02710228
SP
718static void mmap_limit_check(size_t length)
719{
720 static size_t limit = 0;
721 if (!limit) {
722 limit = git_env_ulong("GIT_MMAP_LIMIT", 0);
723 if (!limit)
724 limit = SIZE_MAX;
725 }
726 if (length > limit)
727 die("attempting to mmap %"PRIuMAX" over limit %"PRIuMAX,
728 (uintmax_t)length, (uintmax_t)limit);
729}
730
1570856b
JK
731void *xmmap_gently(void *start, size_t length,
732 int prot, int flags, int fd, off_t offset)
58ecbd5e 733{
02710228
SP
734 void *ret;
735
736 mmap_limit_check(length);
737 ret = mmap(start, length, prot, flags, fd, offset);
58ecbd5e
JN
738 if (ret == MAP_FAILED) {
739 if (!length)
740 return NULL;
7c3ecb32 741 release_pack_memory(length);
58ecbd5e 742 ret = mmap(start, length, prot, flags, fd, offset);
58ecbd5e
JN
743 }
744 return ret;
745}
746
1570856b
JK
747void *xmmap(void *start, size_t length,
748 int prot, int flags, int fd, off_t offset)
749{
750 void *ret = xmmap_gently(start, length, prot, flags, fd, offset);
751 if (ret == MAP_FAILED)
9ca0aaf6 752 die_errno("mmap failed");
1570856b
JK
753 return ret;
754}
755
c9ced051
SP
756void close_pack_windows(struct packed_git *p)
757{
758 while (p->windows) {
759 struct pack_window *w = p->windows;
760
761 if (w->inuse_cnt)
762 die("pack '%s' still has open windows to it",
763 p->pack_name);
764 munmap(w->base, w->len);
765 pack_mapped -= w->len;
766 pack_open_windows--;
767 p->windows = w->next;
768 free(w);
769 }
770}
771
88d0db55
BC
772/*
773 * The LRU pack is the one with the oldest MRU window, preferring packs
774 * with no used windows, or the oldest mtime if it has no windows allocated.
775 */
776static void find_lru_pack(struct packed_git *p, struct packed_git **lru_p, struct pack_window **mru_w, int *accept_windows_inuse)
777{
778 struct pack_window *w, *this_mru_w;
779 int has_windows_inuse = 0;
780
781 /*
782 * Reject this pack if it has windows and the previously selected
783 * one does not. If this pack does not have windows, reject
784 * it if the pack file is newer than the previously selected one.
785 */
786 if (*lru_p && !*mru_w && (p->windows || p->mtime > (*lru_p)->mtime))
787 return;
788
789 for (w = this_mru_w = p->windows; w; w = w->next) {
790 /*
791 * Reject this pack if any of its windows are in use,
792 * but the previously selected pack did not have any
793 * inuse windows. Otherwise, record that this pack
794 * has windows in use.
795 */
796 if (w->inuse_cnt) {
797 if (*accept_windows_inuse)
798 has_windows_inuse = 1;
799 else
800 return;
801 }
802
803 if (w->last_used > this_mru_w->last_used)
804 this_mru_w = w;
805
806 /*
807 * Reject this pack if it has windows that have been
808 * used more recently than the previously selected pack.
809 * If the previously selected pack had windows inuse and
810 * we have not encountered a window in this pack that is
811 * inuse, skip this check since we prefer a pack with no
812 * inuse windows to one that has inuse windows.
813 */
814 if (*mru_w && *accept_windows_inuse == has_windows_inuse &&
815 this_mru_w->last_used > (*mru_w)->last_used)
816 return;
817 }
818
819 /*
820 * Select this pack.
821 */
822 *mru_w = this_mru_w;
823 *lru_p = p;
824 *accept_windows_inuse = has_windows_inuse;
825}
826
827static int close_one_pack(void)
828{
829 struct packed_git *p, *lru_p = NULL;
830 struct pack_window *mru_w = NULL;
831 int accept_windows_inuse = 1;
832
833 for (p = packed_git; p; p = p->next) {
834 if (p->pack_fd == -1)
835 continue;
836 find_lru_pack(p, &lru_p, &mru_w, &accept_windows_inuse);
837 }
838
839 if (lru_p) {
840 close(lru_p->pack_fd);
841 pack_open_fds--;
842 lru_p->pack_fd = -1;
843 return 1;
844 }
845
846 return 0;
847}
848
03e79c88 849void unuse_pack(struct pack_window **w_cursor)
f9253394 850{
03e79c88
SP
851 struct pack_window *w = *w_cursor;
852 if (w) {
853 w->inuse_cnt--;
854 *w_cursor = NULL;
855 }
1f688557
JH
856}
857
fa5fc15d
SP
858void close_pack_index(struct packed_git *p)
859{
860 if (p->index_data) {
861 munmap((void *)p->index_data, p->index_size);
862 p->index_data = NULL;
863 }
864}
865
c74faea1
NP
866/*
867 * This is used by git-repack in case a newly created pack happens to
868 * contain the same set of objects as an existing one. In that case
869 * the resulting file might be different even if its name would be the
870 * same. It is best to close any reference to the old pack before it is
01689909 871 * replaced on disk. Of course no index pointers or windows for given pack
c74faea1
NP
872 * must subsist at this point. If ever objects from this pack are requested
873 * again, the new version of the pack will be reinitialized through
874 * reprepare_packed_git().
875 */
876void free_pack_by_name(const char *pack_name)
877{
878 struct packed_git *p, **pp = &packed_git;
879
880 while (*pp) {
881 p = *pp;
882 if (strcmp(pack_name, p->pack_name) == 0) {
fa3a0c94 883 clear_delta_base_cache();
c74faea1 884 close_pack_windows(p);
c7934306 885 if (p->pack_fd != -1) {
c74faea1 886 close(p->pack_fd);
c7934306
SP
887 pack_open_fds--;
888 }
fa5fc15d 889 close_pack_index(p);
c74faea1
NP
890 free(p->bad_object_sha1);
891 *pp = p->next;
c01f51cc
NTND
892 if (last_found_pack == p)
893 last_found_pack = NULL;
c74faea1
NP
894 free(p);
895 return;
896 }
897 pp = &p->next;
898 }
899}
900
a0788266
JS
901static unsigned int get_max_fd_limit(void)
902{
903#ifdef RLIMIT_NOFILE
491a8dec
JH
904 {
905 struct rlimit lim;
a0788266 906
491a8dec
JH
907 if (!getrlimit(RLIMIT_NOFILE, &lim))
908 return lim.rlim_cur;
909 }
910#endif
911
912#ifdef _SC_OPEN_MAX
913 {
914 long open_max = sysconf(_SC_OPEN_MAX);
915 if (0 < open_max)
916 return open_max;
917 /*
918 * Otherwise, we got -1 for one of the two
919 * reasons:
920 *
921 * (1) sysconf() did not understand _SC_OPEN_MAX
922 * and signaled an error with -1; or
923 * (2) sysconf() said there is no limit.
924 *
925 * We _could_ clear errno before calling sysconf() to
926 * tell these two cases apart and return a huge number
927 * in the latter case to let the caller cap it to a
928 * value that is not so selfish, but letting the
929 * fallback OPEN_MAX codepath take care of these cases
930 * is a lot simpler.
931 */
932 }
933#endif
a0788266 934
491a8dec 935#ifdef OPEN_MAX
a0788266
JS
936 return OPEN_MAX;
937#else
938 return 1; /* see the caller ;-) */
939#endif
940}
941
3cf8b462
SP
942/*
943 * Do not call this directly as this leaks p->pack_fd on error return;
944 * call open_packed_git() instead.
945 */
946static int open_packed_git_1(struct packed_git *p)
1f688557 947{
9bc879c1
SP
948 struct stat st;
949 struct pack_header hdr;
950 unsigned char sha1[20];
951 unsigned char *idx_sha1;
2c039da8 952 long fd_flag;
9bc879c1 953
d079837e
SP
954 if (!p->index_data && open_pack_index(p))
955 return error("packfile %s index unavailable", p->pack_name);
956
c7934306 957 if (!pack_max_fds) {
a0788266 958 unsigned int max_fds = get_max_fd_limit();
c7934306
SP
959
960 /* Save 3 for stdin/stdout/stderr, 22 for work */
961 if (25 < max_fds)
962 pack_max_fds = max_fds - 25;
963 else
964 pack_max_fds = 1;
965 }
966
88d0db55 967 while (pack_max_fds <= pack_open_fds && close_one_pack())
c7934306
SP
968 ; /* nothing */
969
f4e51683 970 p->pack_fd = git_open_noatime(p->pack_name);
9bc879c1 971 if (p->pack_fd < 0 || fstat(p->pack_fd, &st))
072db278 972 return -1;
c7934306 973 pack_open_fds++;
9bc879c1
SP
974
975 /* If we created the struct before we had the pack we lack size. */
bf592c50 976 if (!p->pack_size) {
bf592c50 977 if (!S_ISREG(st.st_mode))
072db278 978 return error("packfile %s not a regular file", p->pack_name);
bf592c50 979 p->pack_size = st.st_size;
9bc879c1 980 } else if (p->pack_size != st.st_size)
072db278 981 return error("packfile %s size changed", p->pack_name);
9bc879c1 982
2c039da8
JH
983 /* We leave these file descriptors open with sliding mmap;
984 * there is no point keeping them open across exec(), though.
985 */
986 fd_flag = fcntl(p->pack_fd, F_GETFD, 0);
987 if (fd_flag < 0)
072db278 988 return error("cannot determine file descriptor flags");
2c039da8
JH
989 fd_flag |= FD_CLOEXEC;
990 if (fcntl(p->pack_fd, F_SETFD, fd_flag) == -1)
072db278 991 return error("cannot set FD_CLOEXEC");
2c039da8 992
9bc879c1 993 /* Verify we recognize this pack file format. */
e6e2bd62 994 if (read_in_full(p->pack_fd, &hdr, sizeof(hdr)) != sizeof(hdr))
072db278 995 return error("file %s is far too short to be a packfile", p->pack_name);
9bc879c1 996 if (hdr.hdr_signature != htonl(PACK_SIGNATURE))
072db278 997 return error("file %s is not a GIT packfile", p->pack_name);
9bc879c1 998 if (!pack_version_ok(hdr.hdr_version))
6e1c2344
RJ
999 return error("packfile %s is version %"PRIu32" and not"
1000 " supported (try upgrading GIT to a newer version)",
9bc879c1
SP
1001 p->pack_name, ntohl(hdr.hdr_version));
1002
1003 /* Verify the pack matches its index. */
57059091 1004 if (p->num_objects != ntohl(hdr.hdr_entries))
6e1c2344
RJ
1005 return error("packfile %s claims to have %"PRIu32" objects"
1006 " while index indicates %"PRIu32" objects",
57059091
NP
1007 p->pack_name, ntohl(hdr.hdr_entries),
1008 p->num_objects);
9bc879c1 1009 if (lseek(p->pack_fd, p->pack_size - sizeof(sha1), SEEK_SET) == -1)
072db278 1010 return error("end of packfile %s is unavailable", p->pack_name);
e6e2bd62 1011 if (read_in_full(p->pack_fd, sha1, sizeof(sha1)) != sizeof(sha1))
072db278 1012 return error("packfile %s signature is unavailable", p->pack_name);
42873078 1013 idx_sha1 = ((unsigned char *)p->index_data) + p->index_size - 40;
9bc879c1 1014 if (hashcmp(sha1, idx_sha1))
072db278
SP
1015 return error("packfile %s does not match index", p->pack_name);
1016 return 0;
9bc879c1
SP
1017}
1018
3cf8b462
SP
1019static int open_packed_git(struct packed_git *p)
1020{
1021 if (!open_packed_git_1(p))
1022 return 0;
1023 if (p->pack_fd != -1) {
1024 close(p->pack_fd);
c7934306 1025 pack_open_fds--;
3cf8b462
SP
1026 p->pack_fd = -1;
1027 }
1028 return -1;
1029}
1030
c4001d92 1031static int in_window(struct pack_window *win, off_t offset)
60bb8b14
SP
1032{
1033 /* We must promise at least 20 bytes (one hash) after the
1034 * offset is available from this window, otherwise the offset
1035 * is not actually in this window and a different window (which
1036 * has that one hash excess) must be used. This is to support
1037 * the object header and delta base parsing routines below.
1038 */
1039 off_t win_off = win->offset;
1040 return win_off <= offset
1041 && (offset + 20) <= (win_off + win->len);
1042}
1043
4b25d091 1044unsigned char *use_pack(struct packed_git *p,
03e79c88 1045 struct pack_window **w_cursor,
c4001d92 1046 off_t offset,
ef49a7a0 1047 unsigned long *left)
9bc879c1 1048{
60bb8b14 1049 struct pack_window *win = *w_cursor;
03e79c88 1050
a9d98a14 1051 /* Since packfiles end in a hash of their content and it's
60bb8b14
SP
1052 * pointless to ask for an offset into the middle of that
1053 * hash, and the in_window function above wouldn't match
1054 * don't allow an offset too close to the end of the file.
1055 */
d131b7af
SP
1056 if (!p->pack_size && p->pack_fd == -1 && open_packed_git(p))
1057 die("packfile %s cannot be accessed", p->pack_name);
60bb8b14
SP
1058 if (offset > (p->pack_size - 20))
1059 die("offset beyond end of packfile (truncated pack?)");
1060
1061 if (!win || !in_window(win, offset)) {
1062 if (win)
1063 win->inuse_cnt--;
1064 for (win = p->windows; win; win = win->next) {
1065 if (in_window(win, offset))
1066 break;
1067 }
1068 if (!win) {
78a28df9 1069 size_t window_align = packed_git_window_size / 2;
dc49cd76 1070 off_t len;
d131b7af
SP
1071
1072 if (p->pack_fd == -1 && open_packed_git(p))
1073 die("packfile %s cannot be accessed", p->pack_name);
1074
60bb8b14 1075 win = xcalloc(1, sizeof(*win));
78a28df9 1076 win->offset = (offset / window_align) * window_align;
dc49cd76
SP
1077 len = p->pack_size - win->offset;
1078 if (len > packed_git_window_size)
1079 len = packed_git_window_size;
1080 win->len = (size_t)len;
60bb8b14 1081 pack_mapped += win->len;
11daf39b 1082 while (packed_git_limit < pack_mapped
7c3ecb32 1083 && unuse_one_window(p))
60bb8b14 1084 ; /* nothing */
c4712e45 1085 win->base = xmmap(NULL, win->len,
60bb8b14
SP
1086 PROT_READ, MAP_PRIVATE,
1087 p->pack_fd, win->offset);
1088 if (win->base == MAP_FAILED)
73b4e4be
SP
1089 die("packfile %s cannot be mapped: %s",
1090 p->pack_name,
1091 strerror(errno));
d131b7af
SP
1092 if (!win->offset && win->len == p->pack_size
1093 && !p->do_not_close) {
1094 close(p->pack_fd);
1095 pack_open_fds--;
1096 p->pack_fd = -1;
1097 }
a53128b6
SP
1098 pack_mmap_calls++;
1099 pack_open_windows++;
1100 if (pack_mapped > peak_pack_mapped)
1101 peak_pack_mapped = pack_mapped;
1102 if (pack_open_windows > peak_pack_open_windows)
1103 peak_pack_open_windows = pack_open_windows;
60bb8b14
SP
1104 win->next = p->windows;
1105 p->windows = win;
1106 }
1f688557 1107 }
03e79c88
SP
1108 if (win != *w_cursor) {
1109 win->last_used = pack_used_ctr++;
1110 win->inuse_cnt++;
1111 *w_cursor = win;
1112 }
60bb8b14 1113 offset -= win->offset;
03e79c88 1114 if (left)
dc49cd76 1115 *left = win->len - xsize_t(offset);
03e79c88 1116 return win->base + offset;
1f688557
JH
1117}
1118
27d69a46
NP
1119static struct packed_git *alloc_packed_git(int extra)
1120{
1121 struct packed_git *p = xmalloc(sizeof(*p) + extra);
1122 memset(p, 0, sizeof(*p));
1123 p->pack_fd = -1;
1124 return p;
1125}
1126
e0500293
JN
1127static void try_to_free_pack_memory(size_t size)
1128{
7c3ecb32 1129 release_pack_memory(size);
e0500293
JN
1130}
1131
48bcc1c3 1132struct packed_git *add_packed_git(const char *path, size_t path_len, int local)
1f688557 1133{
e0500293 1134 static int have_set_try_to_free_routine;
1f688557 1135 struct stat st;
48bcc1c3
JK
1136 size_t alloc;
1137 struct packed_git *p;
1f688557 1138
e0500293
JN
1139 if (!have_set_try_to_free_routine) {
1140 have_set_try_to_free_routine = 1;
1141 set_try_to_free_routine(try_to_free_pack_memory);
1142 }
1143
42873078
NP
1144 /*
1145 * Make sure a corresponding .pack file exists and that
1146 * the index looks sane.
1147 */
48bcc1c3 1148 if (!strip_suffix_mem(path, &path_len, ".idx"))
1f688557 1149 return NULL;
48bcc1c3
JK
1150
1151 /*
1152 * ".pack" is long enough to hold any suffix we're adding (and
1153 * the use xsnprintf double-checks that)
1154 */
1155 alloc = path_len + strlen(".pack") + 1;
1156 p = alloc_packed_git(alloc);
42873078 1157 memcpy(p->pack_name, path, path_len);
8d25931d 1158
48bcc1c3 1159 xsnprintf(p->pack_name + path_len, alloc - path_len, ".keep");
8d25931d
BC
1160 if (!access(p->pack_name, F_OK))
1161 p->pack_keep = 1;
1162
48bcc1c3 1163 xsnprintf(p->pack_name + path_len, alloc - path_len, ".pack");
d079837e 1164 if (stat(p->pack_name, &st) || !S_ISREG(st.st_mode)) {
42873078 1165 free(p);
1f688557
JH
1166 return NULL;
1167 }
42873078 1168
1f688557
JH
1169 /* ok, it looks sane as far as we can check without
1170 * actually mapping the pack file.
1171 */
1f688557 1172 p->pack_size = st.st_size;
9d835df2 1173 p->pack_local = local;
b867092f 1174 p->mtime = st.st_mtime;
42873078
NP
1175 if (path_len < 40 || get_sha1_hex(path + path_len - 40, p->sha1))
1176 hashclr(p->sha1);
1f688557
JH
1177 return p;
1178}
1179
7b64469a 1180struct packed_git *parse_pack_index(unsigned char *sha1, const char *idx_path)
c508df5e 1181{
42873078 1182 const char *path = sha1_pack_name(sha1);
c7ab0ba3
JK
1183 int alloc = strlen(path) + 1;
1184 struct packed_git *p = alloc_packed_git(alloc);
bf592c50 1185
c7ab0ba3 1186 memcpy(p->pack_name, path, alloc); /* includes NUL */
27d69a46 1187 hashcpy(p->sha1, sha1);
42873078
NP
1188 if (check_packed_git_idx(idx_path, p)) {
1189 free(p);
bf592c50 1190 return NULL;
42873078 1191 }
bf592c50 1192
bf592c50
DB
1193 return p;
1194}
1195
1196void install_packed_git(struct packed_git *pack)
1197{
c7934306
SP
1198 if (pack->pack_fd != -1)
1199 pack_open_fds++;
1200
bf592c50
DB
1201 pack->next = packed_git;
1202 packed_git = pack;
1203}
1204
543c5caa
NTND
1205void (*report_garbage)(const char *desc, const char *path);
1206
1207static void report_helper(const struct string_list *list,
1208 int seen_bits, int first, int last)
1209{
1210 const char *msg;
1211 switch (seen_bits) {
1212 case 0:
235e8d59 1213 msg = "no corresponding .idx or .pack";
543c5caa
NTND
1214 break;
1215 case 1:
1216 msg = "no corresponding .idx";
1217 break;
1218 case 2:
1219 msg = "no corresponding .pack";
1220 break;
1221 default:
1222 return;
1223 }
1224 for (; first < last; first++)
1225 report_garbage(msg, list->items[first].string);
1226}
1227
1228static void report_pack_garbage(struct string_list *list)
1229{
1230 int i, baselen = -1, first = 0, seen_bits = 0;
1231
1232 if (!report_garbage)
1233 return;
1234
3383e199 1235 string_list_sort(list);
543c5caa
NTND
1236
1237 for (i = 0; i < list->nr; i++) {
1238 const char *path = list->items[i].string;
1239 if (baselen != -1 &&
1240 strncmp(path, list->items[first].string, baselen)) {
1241 report_helper(list, seen_bits, first, i);
1242 baselen = -1;
1243 seen_bits = 0;
1244 }
1245 if (baselen == -1) {
1246 const char *dot = strrchr(path, '.');
1247 if (!dot) {
1248 report_garbage("garbage found", path);
1249 continue;
1250 }
1251 baselen = dot - path + 1;
1252 first = i;
1253 }
1254 if (!strcmp(path + baselen, "pack"))
1255 seen_bits |= 1;
1256 else if (!strcmp(path + baselen, "idx"))
1257 seen_bits |= 2;
1258 }
1259 report_helper(list, seen_bits, first, list->nr);
1260}
1261
9d835df2 1262static void prepare_packed_git_one(char *objdir, int local)
1f688557 1263{
880fb8de
RS
1264 struct strbuf path = STRBUF_INIT;
1265 size_t dirnamelen;
1f688557
JH
1266 DIR *dir;
1267 struct dirent *de;
543c5caa 1268 struct string_list garbage = STRING_LIST_INIT_DUP;
1f688557 1269
880fb8de
RS
1270 strbuf_addstr(&path, objdir);
1271 strbuf_addstr(&path, "/pack");
1272 dir = opendir(path.buf);
b5b16990 1273 if (!dir) {
26125f6b 1274 if (errno != ENOENT)
bd2afde8 1275 error("unable to open object pack directory: %s: %s",
880fb8de
RS
1276 path.buf, strerror(errno));
1277 strbuf_release(&path);
1f688557 1278 return;
b5b16990 1279 }
880fb8de
RS
1280 strbuf_addch(&path, '/');
1281 dirnamelen = path.len;
1f688557 1282 while ((de = readdir(dir)) != NULL) {
1f688557 1283 struct packed_git *p;
47bf4b0f 1284 size_t base_len;
543c5caa
NTND
1285
1286 if (is_dot_or_dotdot(de->d_name))
9cb18f56
JM
1287 continue;
1288
880fb8de
RS
1289 strbuf_setlen(&path, dirnamelen);
1290 strbuf_addstr(&path, de->d_name);
d90906a9 1291
47bf4b0f
JK
1292 base_len = path.len;
1293 if (strip_suffix_mem(path.buf, &base_len, ".idx")) {
d90906a9
NTND
1294 /* Don't reopen a pack we already have. */
1295 for (p = packed_git; p; p = p->next) {
47bf4b0f
JK
1296 size_t len;
1297 if (strip_suffix(p->pack_name, ".pack", &len) &&
1298 len == base_len &&
1299 !memcmp(p->pack_name, path.buf, len))
d90906a9
NTND
1300 break;
1301 }
1302 if (p == NULL &&
1303 /*
1304 * See if it really is a valid .idx file with
1305 * corresponding .pack file that we can map.
1306 */
880fb8de 1307 (p = add_packed_git(path.buf, path.len, local)) != NULL)
d90906a9 1308 install_packed_git(p);
86f7780c 1309 }
543c5caa
NTND
1310
1311 if (!report_garbage)
1312 continue;
1313
2975c770
JK
1314 if (ends_with(de->d_name, ".idx") ||
1315 ends_with(de->d_name, ".pack") ||
1316 ends_with(de->d_name, ".bitmap") ||
1317 ends_with(de->d_name, ".keep"))
880fb8de 1318 string_list_append(&garbage, path.buf);
543c5caa 1319 else
880fb8de 1320 report_garbage("garbage found", path.buf);
1f688557 1321 }
5b35bcd5 1322 closedir(dir);
543c5caa
NTND
1323 report_pack_garbage(&garbage);
1324 string_list_clear(&garbage, 0);
880fb8de 1325 strbuf_release(&path);
1f688557
JH
1326}
1327
b867092f
JH
1328static int sort_pack(const void *a_, const void *b_)
1329{
1330 struct packed_git *a = *((struct packed_git **)a_);
1331 struct packed_git *b = *((struct packed_git **)b_);
1332 int st;
1333
1334 /*
1335 * Local packs tend to contain objects specific to our
1336 * variant of the project than remote ones. In addition,
1337 * remote ones could be on a network mounted filesystem.
1338 * Favor local ones for these reasons.
1339 */
1340 st = a->pack_local - b->pack_local;
1341 if (st)
1342 return -st;
1343
1344 /*
1345 * Younger packs tend to contain more recent objects,
1346 * and more recent objects tend to get accessed more
1347 * often.
1348 */
1349 if (a->mtime < b->mtime)
1350 return 1;
1351 else if (a->mtime == b->mtime)
1352 return 0;
1353 return -1;
1354}
1355
1356static void rearrange_packed_git(void)
1357{
1358 struct packed_git **ary, *p;
1359 int i, n;
1360
1361 for (n = 0, p = packed_git; p; p = p->next)
1362 n++;
1363 if (n < 2)
1364 return;
1365
1366 /* prepare an array of packed_git for easier sorting */
1367 ary = xcalloc(n, sizeof(struct packed_git *));
1368 for (n = 0, p = packed_git; p; p = p->next)
1369 ary[n++] = p;
1370
1371 qsort(ary, n, sizeof(struct packed_git *), sort_pack);
1372
1373 /* link them back again */
1374 for (i = 0; i < n - 1; i++)
1375 ary[i]->next = ary[i + 1];
1376 ary[n - 1]->next = NULL;
1377 packed_git = ary[0];
1378
1379 free(ary);
1380}
1381
637cdd9d 1382static int prepare_packed_git_run_once = 0;
9a217f2a 1383void prepare_packed_git(void)
1f688557 1384{
d5a63b99 1385 struct alternate_object_database *alt;
1f688557 1386
637cdd9d 1387 if (prepare_packed_git_run_once)
1f688557 1388 return;
9d835df2 1389 prepare_packed_git_one(get_object_directory(), 1);
9a217f2a 1390 prepare_alt_odb();
d5a63b99 1391 for (alt = alt_odb_list; alt; alt = alt->next) {
1494e038 1392 alt->name[-1] = 0;
9d835df2 1393 prepare_packed_git_one(alt->base, 0);
1494e038 1394 alt->name[-1] = '/';
1f688557 1395 }
b867092f 1396 rearrange_packed_git();
637cdd9d
JK
1397 prepare_packed_git_run_once = 1;
1398}
1399
fc04c412 1400void reprepare_packed_git(void)
637cdd9d
JK
1401{
1402 prepare_packed_git_run_once = 0;
1403 prepare_packed_git();
1f688557
JH
1404}
1405
8eca0b47
NP
1406static void mark_bad_packed_object(struct packed_git *p,
1407 const unsigned char *sha1)
1408{
1409 unsigned i;
1410 for (i = 0; i < p->num_bad_objects; i++)
1411 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
1412 return;
1413 p->bad_object_sha1 = xrealloc(p->bad_object_sha1, 20 * (p->num_bad_objects + 1));
1414 hashcpy(p->bad_object_sha1 + 20 * p->num_bad_objects, sha1);
1415 p->num_bad_objects++;
1416}
1417
b6c4cecc 1418static const struct packed_git *has_packed_and_bad(const unsigned char *sha1)
ac939109
NP
1419{
1420 struct packed_git *p;
1421 unsigned i;
1422
1423 for (p = packed_git; p; p = p->next)
1424 for (i = 0; i < p->num_bad_objects; i++)
1425 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
b6c4cecc
JH
1426 return p;
1427 return NULL;
ac939109
NP
1428}
1429
090ea126
NTND
1430/*
1431 * With an in-core object data in "map", rehash it to make sure the
1432 * object name actually matches "sha1" to detect object corruption.
1433 * With "map" == NULL, try reading the object named with "sha1" using
1434 * the streaming interface and rehash it to do the same.
1435 */
1436int check_sha1_signature(const unsigned char *sha1, void *map,
1437 unsigned long size, const char *type)
0fcfd160
LT
1438{
1439 unsigned char real_sha1[20];
090ea126
NTND
1440 enum object_type obj_type;
1441 struct git_istream *st;
1442 git_SHA_CTX c;
1443 char hdr[32];
1444 int hdrlen;
1445
1446 if (map) {
1447 hash_sha1_file(map, size, type, real_sha1);
1448 return hashcmp(sha1, real_sha1) ? -1 : 0;
1449 }
1450
1451 st = open_istream(sha1, &obj_type, &size, NULL);
1452 if (!st)
1453 return -1;
1454
1455 /* Generate the header */
ef1286d3 1456 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(obj_type), size) + 1;
090ea126
NTND
1457
1458 /* Sha1.. */
1459 git_SHA1_Init(&c);
1460 git_SHA1_Update(&c, hdr, hdrlen);
1461 for (;;) {
1462 char buf[1024 * 16];
1463 ssize_t readlen = read_istream(st, buf, sizeof(buf));
1464
f54fac53
JK
1465 if (readlen < 0) {
1466 close_istream(st);
1467 return -1;
1468 }
090ea126
NTND
1469 if (!readlen)
1470 break;
1471 git_SHA1_Update(&c, buf, readlen);
1472 }
1473 git_SHA1_Final(real_sha1, &c);
1474 close_istream(st);
a89fccd2 1475 return hashcmp(sha1, real_sha1) ? -1 : 0;
0fcfd160
LT
1476}
1477
ec73f580 1478int git_open_noatime(const char *name)
44d1c19e
LT
1479{
1480 static int sha1_file_open_flag = O_NOATIME;
44d1c19e 1481
f2e872aa 1482 for (;;) {
dff6f280
CB
1483 int fd;
1484
1485 errno = 0;
1486 fd = open(name, O_RDONLY | sha1_file_open_flag);
44d1c19e 1487 if (fd >= 0)
f2e872aa
SP
1488 return fd;
1489
f2e872aa
SP
1490 /* Might the failure be due to O_NOATIME? */
1491 if (errno != ENOENT && sha1_file_open_flag) {
44d1c19e 1492 sha1_file_open_flag = 0;
f2e872aa
SP
1493 continue;
1494 }
1495
1496 return -1;
44d1c19e 1497 }
44d1c19e
LT
1498}
1499
052fe5ea
JK
1500static int stat_sha1_file(const unsigned char *sha1, struct stat *st)
1501{
052fe5ea
JK
1502 struct alternate_object_database *alt;
1503
30d6c6ea 1504 if (!lstat(sha1_file_name(sha1), st))
052fe5ea
JK
1505 return 0;
1506
1507 prepare_alt_odb();
1508 errno = ENOENT;
1509 for (alt = alt_odb_list; alt; alt = alt->next) {
30d6c6ea 1510 fill_sha1_path(alt->name, sha1);
052fe5ea
JK
1511 if (!lstat(alt->base, st))
1512 return 0;
1513 }
1514
1515 return -1;
1516}
1517
44d1c19e
LT
1518static int open_sha1_file(const unsigned char *sha1)
1519{
1520 int fd;
44d1c19e 1521 struct alternate_object_database *alt;
d6c8a05b 1522 int most_interesting_errno;
44d1c19e 1523
30d6c6ea 1524 fd = git_open_noatime(sha1_file_name(sha1));
44d1c19e
LT
1525 if (fd >= 0)
1526 return fd;
d6c8a05b 1527 most_interesting_errno = errno;
44d1c19e
LT
1528
1529 prepare_alt_odb();
44d1c19e 1530 for (alt = alt_odb_list; alt; alt = alt->next) {
30d6c6ea 1531 fill_sha1_path(alt->name, sha1);
f4e51683 1532 fd = git_open_noatime(alt->base);
44d1c19e
LT
1533 if (fd >= 0)
1534 return fd;
d6c8a05b
JK
1535 if (most_interesting_errno == ENOENT)
1536 most_interesting_errno = errno;
44d1c19e 1537 }
d6c8a05b 1538 errno = most_interesting_errno;
44d1c19e
LT
1539 return -1;
1540}
1541
f0270efd 1542void *map_sha1_file(const unsigned char *sha1, unsigned long *size)
0fcfd160 1543{
0fcfd160 1544 void *map;
144bde78 1545 int fd;
ace1534d 1546
44d1c19e
LT
1547 fd = open_sha1_file(sha1);
1548 map = NULL;
1549 if (fd >= 0) {
1550 struct stat st;
0fcfd160 1551
44d1c19e
LT
1552 if (!fstat(fd, &st)) {
1553 *size = xsize_t(st.st_size);
33e42de0
MM
1554 if (!*size) {
1555 /* mmap() is forbidden on empty files */
1556 error("object file %s is empty", sha1_file_name(sha1));
1557 return NULL;
1558 }
44d1c19e 1559 map = xmmap(NULL, *size, PROT_READ, MAP_PRIVATE, fd, 0);
144bde78 1560 }
44d1c19e 1561 close(fd);
0fcfd160 1562 }
0fcfd160
LT
1563 return map;
1564}
1565
09ded04b
NP
1566unsigned long unpack_object_header_buffer(const unsigned char *buf,
1567 unsigned long len, enum object_type *type, unsigned long *sizep)
c4483576 1568{
ad1ed5ee 1569 unsigned shift;
48fb7deb 1570 unsigned long size, c;
ad1ed5ee
JH
1571 unsigned long used = 0;
1572
1573 c = buf[used++];
1574 *type = (c >> 4) & 7;
1575 size = c & 15;
1576 shift = 4;
1577 while (c & 0x80) {
f630cfda 1578 if (len <= used || bitsizeof(long) <= shift) {
09ded04b 1579 error("bad object header");
ea4f9685
JH
1580 size = used = 0;
1581 break;
09ded04b 1582 }
ad1ed5ee
JH
1583 c = buf[used++];
1584 size += (c & 0x7f) << shift;
1585 shift += 7;
1586 }
1587 *sizep = size;
1588 return used;
1589}
1590
eb4f4076 1591int unpack_sha1_header(git_zstream *stream, unsigned char *map, unsigned long mapsize, void *buffer, unsigned long bufsiz)
ad1ed5ee 1592{
c4483576
LT
1593 /* Get the data stream */
1594 memset(stream, 0, sizeof(*stream));
1595 stream->next_in = map;
1596 stream->avail_in = mapsize;
1597 stream->next_out = buffer;
93821bd9
LT
1598 stream->avail_out = bufsiz;
1599
39c68542 1600 git_inflate_init(stream);
cc5c54e7 1601 return git_inflate(stream, 0);
c4483576
LT
1602}
1603
46f03448
KN
1604static int unpack_sha1_header_to_strbuf(git_zstream *stream, unsigned char *map,
1605 unsigned long mapsize, void *buffer,
1606 unsigned long bufsiz, struct strbuf *header)
1607{
1608 int status;
1609
1610 status = unpack_sha1_header(stream, map, mapsize, buffer, bufsiz);
1611
1612 /*
1613 * Check if entire header is unpacked in the first iteration.
1614 */
1615 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1616 return 0;
1617
1618 /*
1619 * buffer[0..bufsiz] was not large enough. Copy the partial
1620 * result out to header, and then append the result of further
1621 * reading the stream.
1622 */
1623 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1624 stream->next_out = buffer;
1625 stream->avail_out = bufsiz;
1626
1627 do {
1628 status = git_inflate(stream, 0);
1629 strbuf_add(header, buffer, stream->next_out - (unsigned char *)buffer);
1630 if (memchr(buffer, '\0', stream->next_out - (unsigned char *)buffer))
1631 return 0;
1632 stream->next_out = buffer;
1633 stream->avail_out = bufsiz;
1634 } while (status != Z_STREAM_END);
1635 return -1;
1636}
1637
ef49a7a0 1638static void *unpack_sha1_rest(git_zstream *stream, void *buffer, unsigned long size, const unsigned char *sha1)
5180cacc
LT
1639{
1640 int bytes = strlen(buffer) + 1;
3aee68aa 1641 unsigned char *buf = xmallocz(size);
93821bd9 1642 unsigned long n;
7efbff75 1643 int status = Z_OK;
5180cacc 1644
93821bd9
LT
1645 n = stream->total_out - bytes;
1646 if (n > size)
1647 n = size;
1648 memcpy(buf, (char *) buffer + bytes, n);
1649 bytes = n;
456cdf6e
LT
1650 if (bytes <= size) {
1651 /*
1652 * The above condition must be (bytes <= size), not
1653 * (bytes < size). In other words, even though we
ccf5ace0 1654 * expect no more output and set avail_out to zero,
456cdf6e
LT
1655 * the input zlib stream may have bytes that express
1656 * "this concludes the stream", and we *do* want to
1657 * eat that input.
1658 *
1659 * Otherwise we would not be able to test that we
1660 * consumed all the input to reach the expected size;
1661 * we also want to check that zlib tells us that all
1662 * went well with status == Z_STREAM_END at the end.
1663 */
5180cacc
LT
1664 stream->next_out = buf + bytes;
1665 stream->avail_out = size - bytes;
7efbff75 1666 while (status == Z_OK)
39c68542 1667 status = git_inflate(stream, Z_FINISH);
5180cacc 1668 }
456cdf6e 1669 if (status == Z_STREAM_END && !stream->avail_in) {
39c68542 1670 git_inflate_end(stream);
7efbff75
JH
1671 return buf;
1672 }
1673
1674 if (status < 0)
1675 error("corrupt loose object '%s'", sha1_to_hex(sha1));
1676 else if (stream->avail_in)
1677 error("garbage at end of loose object '%s'",
1678 sha1_to_hex(sha1));
1679 free(buf);
1680 return NULL;
5180cacc
LT
1681}
1682
1683/*
1684 * We used to just use "sscanf()", but that's actually way
1685 * too permissive for what we want to check. So do an anal
1686 * object header parse by hand.
1687 */
46f03448
KN
1688static int parse_sha1_header_extended(const char *hdr, struct object_info *oi,
1689 unsigned int flags)
5180cacc 1690{
46f03448 1691 const char *type_buf = hdr;
5180cacc 1692 unsigned long size;
46f03448 1693 int type, type_len = 0;
5180cacc
LT
1694
1695 /*
46f03448 1696 * The type can be of any size but is followed by
21666f1a 1697 * a space.
5180cacc 1698 */
5180cacc
LT
1699 for (;;) {
1700 char c = *hdr++;
1701 if (c == ' ')
1702 break;
46f03448 1703 type_len++;
5180cacc 1704 }
46f03448
KN
1705
1706 type = type_from_string_gently(type_buf, type_len, 1);
1707 if (oi->typename)
1708 strbuf_add(oi->typename, type_buf, type_len);
1709 /*
1710 * Set type to 0 if its an unknown object and
1711 * we're obtaining the type using '--allow-unkown-type'
1712 * option.
1713 */
1714 if ((flags & LOOKUP_UNKNOWN_OBJECT) && (type < 0))
1715 type = 0;
1716 else if (type < 0)
1717 die("invalid object type");
1718 if (oi->typep)
1719 *oi->typep = type;
5180cacc
LT
1720
1721 /*
1722 * The length must follow immediately, and be in canonical
1723 * decimal format (ie "010" is not valid).
1724 */
1725 size = *hdr++ - '0';
1726 if (size > 9)
1727 return -1;
1728 if (size) {
1729 for (;;) {
1730 unsigned long c = *hdr - '0';
1731 if (c > 9)
1732 break;
1733 hdr++;
1734 size = size * 10 + c;
1735 }
1736 }
46f03448
KN
1737
1738 if (oi->sizep)
1739 *oi->sizep = size;
5180cacc
LT
1740
1741 /*
1742 * The length must be followed by a zero byte
1743 */
46f03448
KN
1744 return *hdr ? -1 : type;
1745}
1746
1747int parse_sha1_header(const char *hdr, unsigned long *sizep)
1748{
1749 struct object_info oi;
1750
1751 oi.sizep = sizep;
1752 oi.typename = NULL;
1753 oi.typep = NULL;
1754 return parse_sha1_header_extended(hdr, &oi, LOOKUP_REPLACE_OBJECT);
5180cacc
LT
1755}
1756
7efbff75 1757static void *unpack_sha1_file(void *map, unsigned long mapsize, enum object_type *type, unsigned long *size, const unsigned char *sha1)
0fcfd160 1758{
5180cacc 1759 int ret;
ef49a7a0 1760 git_zstream stream;
5180cacc 1761 char hdr[8192];
0fcfd160 1762
5180cacc 1763 ret = unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr));
21666f1a 1764 if (ret < Z_OK || (*type = parse_sha1_header(hdr, size)) < 0)
0fcfd160
LT
1765 return NULL;
1766
7efbff75 1767 return unpack_sha1_rest(&stream, hdr, *size, sha1);
0fcfd160
LT
1768}
1769
54dab52a
NP
1770unsigned long get_size_from_delta(struct packed_git *p,
1771 struct pack_window **w_curs,
1772 off_t curpos)
1773{
1774 const unsigned char *data;
1775 unsigned char delta_head[20], *in;
ef49a7a0 1776 git_zstream stream;
54dab52a
NP
1777 int st;
1778
1779 memset(&stream, 0, sizeof(stream));
1780 stream.next_out = delta_head;
1781 stream.avail_out = sizeof(delta_head);
1782
39c68542 1783 git_inflate_init(&stream);
54dab52a
NP
1784 do {
1785 in = use_pack(p, w_curs, curpos, &stream.avail_in);
1786 stream.next_in = in;
39c68542 1787 st = git_inflate(&stream, Z_FINISH);
54dab52a
NP
1788 curpos += stream.next_in - in;
1789 } while ((st == Z_OK || st == Z_BUF_ERROR) &&
1790 stream.total_out < sizeof(delta_head));
39c68542 1791 git_inflate_end(&stream);
3d77d877
NP
1792 if ((st != Z_STREAM_END) && stream.total_out != sizeof(delta_head)) {
1793 error("delta data unpack-initial failed");
1794 return 0;
1795 }
54dab52a
NP
1796
1797 /* Examine the initial part of the delta to figure out
1798 * the result size.
1799 */
1800 data = delta_head;
1801
1802 /* ignore base size */
1803 get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1804
1805 /* Read the result size */
1806 return get_delta_hdr_size(&data, delta_head+sizeof(delta_head));
1807}
1808
c4001d92 1809static off_t get_delta_base(struct packed_git *p,
03e79c88 1810 struct pack_window **w_curs,
c4001d92 1811 off_t *curpos,
21666f1a 1812 enum object_type type,
c4001d92 1813 off_t delta_obj_offset)
eb32d236 1814{
2b87c45b 1815 unsigned char *base_info = use_pack(p, w_curs, *curpos, NULL);
c4001d92 1816 off_t base_offset;
eb32d236 1817
8d8a4ea5
SP
1818 /* use_pack() assured us we have [base_info, base_info + 20)
1819 * as a range that we can look at without walking off the
1820 * end of the mapped window. Its actually the hash size
1821 * that is assured. An OFS_DELTA longer than the hash size
1822 * is stupid, as then a REF_DELTA would be smaller to store.
1823 */
21666f1a 1824 if (type == OBJ_OFS_DELTA) {
eb32d236
NP
1825 unsigned used = 0;
1826 unsigned char c = base_info[used++];
1827 base_offset = c & 127;
1828 while (c & 128) {
1829 base_offset += 1;
8723f216 1830 if (!base_offset || MSB(base_offset, 7))
8eca0b47 1831 return 0; /* overflow */
eb32d236
NP
1832 c = base_info[used++];
1833 base_offset = (base_offset << 7) + (c & 127);
1834 }
1835 base_offset = delta_obj_offset - base_offset;
d8f32556 1836 if (base_offset <= 0 || base_offset >= delta_obj_offset)
8eca0b47 1837 return 0; /* out of bound */
2b87c45b 1838 *curpos += used;
21666f1a 1839 } else if (type == OBJ_REF_DELTA) {
eb32d236
NP
1840 /* The base entry _must_ be in the same pack */
1841 base_offset = find_pack_entry_one(base_info, p);
2b87c45b 1842 *curpos += 20;
eb32d236
NP
1843 } else
1844 die("I am totally screwed");
2b87c45b 1845 return base_offset;
eb32d236
NP
1846}
1847
5d642e75
JK
1848/*
1849 * Like get_delta_base above, but we return the sha1 instead of the pack
1850 * offset. This means it is cheaper for REF deltas (we do not have to do
1851 * the final object lookup), but more expensive for OFS deltas (we
1852 * have to load the revidx to convert the offset back into a sha1).
1853 */
1854static const unsigned char *get_delta_base_sha1(struct packed_git *p,
1855 struct pack_window **w_curs,
1856 off_t curpos,
1857 enum object_type type,
1858 off_t delta_obj_offset)
1859{
1860 if (type == OBJ_REF_DELTA) {
1861 unsigned char *base = use_pack(p, w_curs, curpos, NULL);
1862 return base;
1863 } else if (type == OBJ_OFS_DELTA) {
1864 struct revindex_entry *revidx;
1865 off_t base_offset = get_delta_base(p, w_curs, &curpos,
1866 type, delta_obj_offset);
1867
1868 if (!base_offset)
1869 return NULL;
1870
1871 revidx = find_pack_revindex(p, base_offset);
1872 if (!revidx)
1873 return NULL;
1874
1875 return nth_packed_object_sha1(p, revidx->nr);
1876 } else
1877 return NULL;
1878}
1879
f8c8abc5
JH
1880int unpack_object_header(struct packed_git *p,
1881 struct pack_window **w_curs,
1882 off_t *curpos,
1883 unsigned long *sizep)
a733cb60 1884{
03e79c88 1885 unsigned char *base;
ef49a7a0 1886 unsigned long left;
ad1ed5ee 1887 unsigned long used;
2b87c45b 1888 enum object_type type;
a733cb60 1889
8d8a4ea5 1890 /* use_pack() assures us we have [base, base + 20) available
0353a0c4 1891 * as a range that we can look at. (Its actually the hash
3dff5379 1892 * size that is assured.) With our object header encoding
8d8a4ea5
SP
1893 * the maximum deflated object size is 2^137, which is just
1894 * insane, so we know won't exceed what we have been given.
1895 */
2b87c45b 1896 base = use_pack(p, w_curs, *curpos, &left);
09ded04b
NP
1897 used = unpack_object_header_buffer(base, left, &type, sizep);
1898 if (!used) {
1899 type = OBJ_BAD;
1900 } else
1901 *curpos += used;
ad1ed5ee 1902
2b87c45b 1903 return type;
a733cb60
LT
1904}
1905
790d96c0
TR
1906static int retry_bad_packed_offset(struct packed_git *p, off_t obj_offset)
1907{
1908 int type;
1909 struct revindex_entry *revidx;
1910 const unsigned char *sha1;
1911 revidx = find_pack_revindex(p, obj_offset);
1912 if (!revidx)
1913 return OBJ_BAD;
1914 sha1 = nth_packed_object_sha1(p, revidx->nr);
1915 mark_bad_packed_object(p, sha1);
1916 type = sha1_object_info(sha1, NULL);
1917 if (type <= OBJ_NONE)
1918 return OBJ_BAD;
1919 return type;
1920}
1921
790d96c0
TR
1922#define POI_STACK_PREALLOC 64
1923
90191d37
JK
1924static enum object_type packed_to_object_type(struct packed_git *p,
1925 off_t obj_offset,
1926 enum object_type type,
1927 struct pack_window **w_curs,
1928 off_t curpos)
1f688557 1929{
790d96c0
TR
1930 off_t small_poi_stack[POI_STACK_PREALLOC];
1931 off_t *poi_stack = small_poi_stack;
1932 int poi_stack_nr = 0, poi_stack_alloc = POI_STACK_PREALLOC;
5db47c2b 1933
790d96c0
TR
1934 while (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
1935 off_t base_offset;
90191d37 1936 unsigned long size;
790d96c0
TR
1937 /* Push the object we're going to leave behind */
1938 if (poi_stack_nr >= poi_stack_alloc && poi_stack == small_poi_stack) {
1939 poi_stack_alloc = alloc_nr(poi_stack_nr);
1940 poi_stack = xmalloc(sizeof(off_t)*poi_stack_alloc);
1941 memcpy(poi_stack, small_poi_stack, sizeof(off_t)*poi_stack_nr);
1942 } else {
1943 ALLOC_GROW(poi_stack, poi_stack_nr+1, poi_stack_alloc);
1944 }
1945 poi_stack[poi_stack_nr++] = obj_offset;
1946 /* If parsing the base offset fails, just unwind */
90191d37 1947 base_offset = get_delta_base(p, w_curs, &curpos, type, obj_offset);
790d96c0
TR
1948 if (!base_offset)
1949 goto unwind;
1950 curpos = obj_offset = base_offset;
90191d37 1951 type = unpack_object_header(p, w_curs, &curpos, &size);
790d96c0
TR
1952 if (type <= OBJ_NONE) {
1953 /* If getting the base itself fails, we first
1954 * retry the base, otherwise unwind */
1955 type = retry_bad_packed_offset(p, base_offset);
1956 if (type > OBJ_NONE)
1957 goto out;
1958 goto unwind;
1959 }
1960 }
1961
21666f1a 1962 switch (type) {
790d96c0 1963 case OBJ_BAD:
a733cb60 1964 case OBJ_COMMIT:
a733cb60 1965 case OBJ_TREE:
a733cb60 1966 case OBJ_BLOB:
a733cb60 1967 case OBJ_TAG:
a69d0943 1968 break;
1f688557 1969 default:
3d77d877
NP
1970 error("unknown object type %i at offset %"PRIuMAX" in %s",
1971 type, (uintmax_t)obj_offset, p->pack_name);
1972 type = OBJ_BAD;
1f688557 1973 }
790d96c0
TR
1974
1975out:
1976 if (poi_stack != small_poi_stack)
1977 free(poi_stack);
21666f1a 1978 return type;
790d96c0
TR
1979
1980unwind:
1981 while (poi_stack_nr) {
1982 obj_offset = poi_stack[--poi_stack_nr];
1983 type = retry_bad_packed_offset(p, obj_offset);
1984 if (type > OBJ_NONE)
1985 goto out;
1986 }
1987 type = OBJ_BAD;
1988 goto out;
1f688557
JH
1989}
1990
90191d37 1991static int packed_object_info(struct packed_git *p, off_t obj_offset,
23c339c0 1992 struct object_info *oi)
90191d37
JK
1993{
1994 struct pack_window *w_curs = NULL;
1995 unsigned long size;
1996 off_t curpos = obj_offset;
1997 enum object_type type;
1998
412916ee
JK
1999 /*
2000 * We always get the representation type, but only convert it to
2001 * a "real" type later if the caller is interested.
2002 */
90191d37
JK
2003 type = unpack_object_header(p, &w_curs, &curpos, &size);
2004
23c339c0 2005 if (oi->sizep) {
90191d37
JK
2006 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2007 off_t tmp_pos = curpos;
2008 off_t base_offset = get_delta_base(p, &w_curs, &tmp_pos,
2009 type, obj_offset);
2010 if (!base_offset) {
2011 type = OBJ_BAD;
2012 goto out;
2013 }
23c339c0
JK
2014 *oi->sizep = get_size_from_delta(p, &w_curs, tmp_pos);
2015 if (*oi->sizep == 0) {
90191d37
JK
2016 type = OBJ_BAD;
2017 goto out;
2018 }
2019 } else {
23c339c0 2020 *oi->sizep = size;
90191d37
JK
2021 }
2022 }
2023
23c339c0 2024 if (oi->disk_sizep) {
90191d37 2025 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
23c339c0 2026 *oi->disk_sizep = revidx[1].offset - obj_offset;
90191d37
JK
2027 }
2028
23c339c0
JK
2029 if (oi->typep) {
2030 *oi->typep = packed_to_object_type(p, obj_offset, type, &w_curs, curpos);
2031 if (*oi->typep < 0) {
412916ee
JK
2032 type = OBJ_BAD;
2033 goto out;
2034 }
2035 }
90191d37 2036
5d642e75
JK
2037 if (oi->delta_base_sha1) {
2038 if (type == OBJ_OFS_DELTA || type == OBJ_REF_DELTA) {
2039 const unsigned char *base;
2040
2041 base = get_delta_base_sha1(p, &w_curs, curpos,
2042 type, obj_offset);
2043 if (!base) {
2044 type = OBJ_BAD;
2045 goto out;
2046 }
2047
2048 hashcpy(oi->delta_base_sha1, base);
2049 } else
2050 hashclr(oi->delta_base_sha1);
2051 }
2052
90191d37
JK
2053out:
2054 unuse_pack(&w_curs);
2055 return type;
2056}
2057
eb950c19 2058static void *unpack_compressed_entry(struct packed_git *p,
03e79c88 2059 struct pack_window **w_curs,
c4001d92 2060 off_t curpos,
eb950c19 2061 unsigned long size)
de530aaa
SP
2062{
2063 int st;
ef49a7a0 2064 git_zstream stream;
079afb18 2065 unsigned char *buffer, *in;
de530aaa 2066
735efde8
NTND
2067 buffer = xmallocz_gently(size);
2068 if (!buffer)
2069 return NULL;
de530aaa 2070 memset(&stream, 0, sizeof(stream));
de530aaa 2071 stream.next_out = buffer;
39eea7bd 2072 stream.avail_out = size + 1;
de530aaa 2073
39c68542 2074 git_inflate_init(&stream);
079afb18 2075 do {
2b87c45b 2076 in = use_pack(p, w_curs, curpos, &stream.avail_in);
079afb18 2077 stream.next_in = in;
39c68542 2078 st = git_inflate(&stream, Z_FINISH);
39eea7bd
JH
2079 if (!stream.avail_out)
2080 break; /* the payload is larger than it should be */
2b87c45b 2081 curpos += stream.next_in - in;
079afb18 2082 } while (st == Z_OK || st == Z_BUF_ERROR);
39c68542 2083 git_inflate_end(&stream);
de530aaa
SP
2084 if ((st != Z_STREAM_END) || stream.total_out != size) {
2085 free(buffer);
2086 return NULL;
2087 }
2088
2089 return buffer;
2090}
2091
e5e01619
LT
2092#define MAX_DELTA_CACHE (256)
2093
18bdec11 2094static size_t delta_base_cached;
5e08ecbf
NP
2095
2096static struct delta_base_cache_lru_list {
2097 struct delta_base_cache_lru_list *prev;
2098 struct delta_base_cache_lru_list *next;
2099} delta_base_cache_lru = { &delta_base_cache_lru, &delta_base_cache_lru };
2100
e5e01619 2101static struct delta_base_cache_entry {
5e08ecbf
NP
2102 struct delta_base_cache_lru_list lru;
2103 void *data;
e5e01619
LT
2104 struct packed_git *p;
2105 off_t base_offset;
2106 unsigned long size;
e5e01619
LT
2107 enum object_type type;
2108} delta_base_cache[MAX_DELTA_CACHE];
2109
2110static unsigned long pack_entry_hash(struct packed_git *p, off_t base_offset)
2111{
2112 unsigned long hash;
2113
2114 hash = (unsigned long)p + (unsigned long)base_offset;
2115 hash += (hash >> 8) + (hash >> 16);
3358004a 2116 return hash % MAX_DELTA_CACHE;
e5e01619
LT
2117}
2118
84dd81c1
TR
2119static struct delta_base_cache_entry *
2120get_delta_base_cache_entry(struct packed_git *p, off_t base_offset)
5266d369
JH
2121{
2122 unsigned long hash = pack_entry_hash(p, base_offset);
84dd81c1
TR
2123 return delta_base_cache + hash;
2124}
2125
2126static int eq_delta_base_cache_entry(struct delta_base_cache_entry *ent,
2127 struct packed_git *p, off_t base_offset)
2128{
5266d369
JH
2129 return (ent->data && ent->p == p && ent->base_offset == base_offset);
2130}
2131
84dd81c1
TR
2132static int in_delta_base_cache(struct packed_git *p, off_t base_offset)
2133{
2134 struct delta_base_cache_entry *ent;
2135 ent = get_delta_base_cache_entry(p, base_offset);
2136 return eq_delta_base_cache_entry(ent, p, base_offset);
2137}
2138
2139static void clear_delta_base_cache_entry(struct delta_base_cache_entry *ent)
2140{
2141 ent->data = NULL;
2142 ent->lru.next->prev = ent->lru.prev;
2143 ent->lru.prev->next = ent->lru.next;
2144 delta_base_cached -= ent->size;
2145}
2146
62f255ad 2147static void *cache_or_unpack_entry(struct packed_git *p, off_t base_offset,
a0cba108 2148 unsigned long *base_size, enum object_type *type, int keep_cache)
62f255ad 2149{
84dd81c1 2150 struct delta_base_cache_entry *ent;
e5e01619 2151 void *ret;
e5e01619 2152
84dd81c1
TR
2153 ent = get_delta_base_cache_entry(p, base_offset);
2154
2155 if (!eq_delta_base_cache_entry(ent, p, base_offset))
749bc58c 2156 return unpack_entry(p, base_offset, type, base_size);
e5e01619 2157
84dd81c1
TR
2158 ret = ent->data;
2159
2160 if (!keep_cache)
2161 clear_delta_base_cache_entry(ent);
2162 else
182af834 2163 ret = xmemdupz(ent->data, ent->size);
e5e01619
LT
2164 *type = ent->type;
2165 *base_size = ent->size;
2166 return ret;
62f255ad
LT
2167}
2168
18bdec11
SP
2169static inline void release_delta_base_cache(struct delta_base_cache_entry *ent)
2170{
2171 if (ent->data) {
2172 free(ent->data);
2173 ent->data = NULL;
5e08ecbf
NP
2174 ent->lru.next->prev = ent->lru.prev;
2175 ent->lru.prev->next = ent->lru.next;
18bdec11
SP
2176 delta_base_cached -= ent->size;
2177 }
2178}
2179
3d20c636
SP
2180void clear_delta_base_cache(void)
2181{
2182 unsigned long p;
2183 for (p = 0; p < MAX_DELTA_CACHE; p++)
2184 release_delta_base_cache(&delta_base_cache[p]);
2185}
2186
62f255ad
LT
2187static void add_delta_base_cache(struct packed_git *p, off_t base_offset,
2188 void *base, unsigned long base_size, enum object_type type)
2189{
5e08ecbf 2190 unsigned long hash = pack_entry_hash(p, base_offset);
e5e01619 2191 struct delta_base_cache_entry *ent = delta_base_cache + hash;
5e08ecbf 2192 struct delta_base_cache_lru_list *lru;
e5e01619 2193
18bdec11
SP
2194 release_delta_base_cache(ent);
2195 delta_base_cached += base_size;
5e08ecbf
NP
2196
2197 for (lru = delta_base_cache_lru.next;
2198 delta_base_cached > delta_base_cache_limit
2199 && lru != &delta_base_cache_lru;
2200 lru = lru->next) {
2201 struct delta_base_cache_entry *f = (void *)lru;
18bdec11
SP
2202 if (f->type == OBJ_BLOB)
2203 release_delta_base_cache(f);
2204 }
5e08ecbf
NP
2205 for (lru = delta_base_cache_lru.next;
2206 delta_base_cached > delta_base_cache_limit
2207 && lru != &delta_base_cache_lru;
2208 lru = lru->next) {
2209 struct delta_base_cache_entry *f = (void *)lru;
2210 release_delta_base_cache(f);
2211 }
18bdec11 2212
e5e01619
LT
2213 ent->p = p;
2214 ent->base_offset = base_offset;
2215 ent->type = type;
2216 ent->data = base;
2217 ent->size = base_size;
5e08ecbf
NP
2218 ent->lru.next = &delta_base_cache_lru;
2219 ent->lru.prev = delta_base_cache_lru.prev;
2220 delta_base_cache_lru.prev->next = &ent->lru;
2221 delta_base_cache_lru.prev = &ent->lru;
62f255ad
LT
2222}
2223
c2c5b270
CC
2224static void *read_object(const unsigned char *sha1, enum object_type *type,
2225 unsigned long *size);
2226
5f44324d
JH
2227static void write_pack_access_log(struct packed_git *p, off_t obj_offset)
2228{
67dc598e
KB
2229 static struct trace_key pack_access = TRACE_KEY_INIT(PACK_ACCESS);
2230 trace_printf_key(&pack_access, "%s %"PRIuMAX"\n",
2231 p->pack_name, (uintmax_t)obj_offset);
5f44324d
JH
2232}
2233
0e8189e2
NP
2234int do_check_packed_object_crc;
2235
abe601bb
TR
2236#define UNPACK_ENTRY_STACK_PREALLOC 64
2237struct unpack_entry_stack_ent {
2238 off_t obj_offset;
2239 off_t curpos;
2240 unsigned long size;
2241};
2242
c4001d92 2243void *unpack_entry(struct packed_git *p, off_t obj_offset,
abe601bb 2244 enum object_type *final_type, unsigned long *final_size)
1f688557 2245{
03e79c88 2246 struct pack_window *w_curs = NULL;
c4001d92 2247 off_t curpos = obj_offset;
abe601bb
TR
2248 void *data = NULL;
2249 unsigned long size;
2250 enum object_type type;
2251 struct unpack_entry_stack_ent small_delta_stack[UNPACK_ENTRY_STACK_PREALLOC];
2252 struct unpack_entry_stack_ent *delta_stack = small_delta_stack;
2253 int delta_stack_nr = 0, delta_stack_alloc = UNPACK_ENTRY_STACK_PREALLOC;
2254 int base_from_cache = 0;
1f688557 2255
67dc598e 2256 write_pack_access_log(p, obj_offset);
5f44324d 2257
abe601bb
TR
2258 /* PHASE 1: drill down to the innermost base object */
2259 for (;;) {
2260 off_t base_offset;
2261 int i;
2262 struct delta_base_cache_entry *ent;
2263
77965f8b
NTND
2264 ent = get_delta_base_cache_entry(p, curpos);
2265 if (eq_delta_base_cache_entry(ent, p, curpos)) {
2266 type = ent->type;
2267 data = ent->data;
2268 size = ent->size;
2269 clear_delta_base_cache_entry(ent);
2270 base_from_cache = 1;
2271 break;
2272 }
2273
abe601bb
TR
2274 if (do_check_packed_object_crc && p->index_version > 1) {
2275 struct revindex_entry *revidx = find_pack_revindex(p, obj_offset);
2276 unsigned long len = revidx[1].offset - obj_offset;
2277 if (check_pack_crc(p, &w_curs, obj_offset, len, revidx->nr)) {
2278 const unsigned char *sha1 =
2279 nth_packed_object_sha1(p, revidx->nr);
2280 error("bad packed object CRC for %s",
2281 sha1_to_hex(sha1));
2282 mark_bad_packed_object(p, sha1);
2283 unuse_pack(&w_curs);
2284 return NULL;
2285 }
2286 }
2287
abe601bb
TR
2288 type = unpack_object_header(p, &w_curs, &curpos, &size);
2289 if (type != OBJ_OFS_DELTA && type != OBJ_REF_DELTA)
2290 break;
2291
2292 base_offset = get_delta_base(p, &w_curs, &curpos, type, obj_offset);
2293 if (!base_offset) {
2294 error("failed to validate delta base reference "
2295 "at offset %"PRIuMAX" from %s",
2296 (uintmax_t)curpos, p->pack_name);
2297 /* bail to phase 2, in hopes of recovery */
2298 data = NULL;
2299 break;
2300 }
2301
2302 /* push object, proceed to base */
2303 if (delta_stack_nr >= delta_stack_alloc
2304 && delta_stack == small_delta_stack) {
2305 delta_stack_alloc = alloc_nr(delta_stack_nr);
2306 delta_stack = xmalloc(sizeof(*delta_stack)*delta_stack_alloc);
2307 memcpy(delta_stack, small_delta_stack,
2308 sizeof(*delta_stack)*delta_stack_nr);
2309 } else {
2310 ALLOC_GROW(delta_stack, delta_stack_nr+1, delta_stack_alloc);
2311 }
2312 i = delta_stack_nr++;
2313 delta_stack[i].obj_offset = obj_offset;
2314 delta_stack[i].curpos = curpos;
2315 delta_stack[i].size = size;
2316
2317 curpos = obj_offset = base_offset;
0e8189e2
NP
2318 }
2319
abe601bb
TR
2320 /* PHASE 2: handle the base */
2321 switch (type) {
eb32d236
NP
2322 case OBJ_OFS_DELTA:
2323 case OBJ_REF_DELTA:
abe601bb
TR
2324 if (data)
2325 die("BUG in unpack_entry: left loop at a valid delta");
4d703a1a 2326 break;
a733cb60 2327 case OBJ_COMMIT:
a733cb60 2328 case OBJ_TREE:
a733cb60 2329 case OBJ_BLOB:
a733cb60 2330 case OBJ_TAG:
abe601bb
TR
2331 if (!base_from_cache)
2332 data = unpack_compressed_entry(p, &w_curs, curpos, size);
4d703a1a 2333 break;
1f688557 2334 default:
8eca0b47
NP
2335 data = NULL;
2336 error("unknown object type %i at offset %"PRIuMAX" in %s",
abe601bb
TR
2337 type, (uintmax_t)obj_offset, p->pack_name);
2338 }
2339
2340 /* PHASE 3: apply deltas in order */
2341
2342 /* invariants:
2343 * 'data' holds the base data, or NULL if there was corruption
2344 */
2345 while (delta_stack_nr) {
2346 void *delta_data;
2347 void *base = data;
2348 unsigned long delta_size, base_size = size;
2349 int i;
2350
2351 data = NULL;
2352
2353 if (base)
2354 add_delta_base_cache(p, obj_offset, base, base_size, type);
2355
2356 if (!base) {
2357 /*
2358 * We're probably in deep shit, but let's try to fetch
2359 * the required base anyway from another pack or loose.
2360 * This is costly but should happen only in the presence
2361 * of a corrupted pack, and is better than failing outright.
2362 */
2363 struct revindex_entry *revidx;
2364 const unsigned char *base_sha1;
2365 revidx = find_pack_revindex(p, obj_offset);
2366 if (revidx) {
2367 base_sha1 = nth_packed_object_sha1(p, revidx->nr);
2368 error("failed to read delta base object %s"
2369 " at offset %"PRIuMAX" from %s",
2370 sha1_to_hex(base_sha1), (uintmax_t)obj_offset,
2371 p->pack_name);
2372 mark_bad_packed_object(p, base_sha1);
2373 base = read_object(base_sha1, &type, &base_size);
2374 }
2375 }
2376
2377 i = --delta_stack_nr;
2378 obj_offset = delta_stack[i].obj_offset;
2379 curpos = delta_stack[i].curpos;
2380 delta_size = delta_stack[i].size;
2381
2382 if (!base)
2383 continue;
2384
2385 delta_data = unpack_compressed_entry(p, &w_curs, curpos, delta_size);
2386
2387 if (!delta_data) {
2388 error("failed to unpack compressed delta "
2389 "at offset %"PRIuMAX" from %s",
2390 (uintmax_t)curpos, p->pack_name);
abe601bb
TR
2391 data = NULL;
2392 continue;
2393 }
2394
2395 data = patch_delta(base, base_size,
2396 delta_data, delta_size,
2397 &size);
1ee886c1
JK
2398
2399 /*
2400 * We could not apply the delta; warn the user, but keep going.
2401 * Our failure will be noticed either in the next iteration of
2402 * the loop, or if this is the final delta, in the caller when
2403 * we return NULL. Those code paths will take care of making
2404 * a more explicit warning and retrying with another copy of
2405 * the object.
2406 */
abe601bb 2407 if (!data)
1ee886c1 2408 error("failed to apply delta");
abe601bb 2409
4b8f772c 2410 free(delta_data);
1f688557 2411 }
abe601bb
TR
2412
2413 *final_type = type;
2414 *final_size = size;
2415
03e79c88 2416 unuse_pack(&w_curs);
019d1e65
NTND
2417
2418 if (delta_stack != small_delta_stack)
2419 free(delta_stack);
2420
21666f1a 2421 return data;
1f688557
JH
2422}
2423
d079837e 2424const unsigned char *nth_packed_object_sha1(struct packed_git *p,
d72308e0 2425 uint32_t n)
9a217f2a 2426{
42873078 2427 const unsigned char *index = p->index_data;
d079837e
SP
2428 if (!index) {
2429 if (open_pack_index(p))
2430 return NULL;
2431 index = p->index_data;
2432 }
57059091 2433 if (n >= p->num_objects)
d72308e0 2434 return NULL;
74e34e1f
NP
2435 index += 4 * 256;
2436 if (p->index_version == 1) {
2437 return index + 24 * n + 4;
2438 } else {
2439 index += 8;
2440 return index + 20 * n;
2441 }
2442}
2443
99093238 2444off_t nth_packed_object_offset(const struct packed_git *p, uint32_t n)
74e34e1f
NP
2445{
2446 const unsigned char *index = p->index_data;
2447 index += 4 * 256;
2448 if (p->index_version == 1) {
2449 return ntohl(*((uint32_t *)(index + 24 * n)));
2450 } else {
2451 uint32_t off;
2452 index += 8 + p->num_objects * (20 + 4);
2453 off = ntohl(*((uint32_t *)(index + 4 * n)));
2454 if (!(off & 0x80000000))
2455 return off;
2456 index += p->num_objects * 4 + (off & 0x7fffffff) * 8;
2457 return (((uint64_t)ntohl(*((uint32_t *)(index + 0)))) << 32) |
2458 ntohl(*((uint32_t *)(index + 4)));
2459 }
9a217f2a
JH
2460}
2461
c4001d92 2462off_t find_pack_entry_one(const unsigned char *sha1,
43057304 2463 struct packed_git *p)
1f688557 2464{
42873078 2465 const uint32_t *level1_ofs = p->index_data;
42873078 2466 const unsigned char *index = p->index_data;
628522ec
JH
2467 unsigned hi, lo, stride;
2468 static int use_lookup = -1;
2469 static int debug_lookup = -1;
2470
2471 if (debug_lookup < 0)
2472 debug_lookup = !!getenv("GIT_DEBUG_LOOKUP");
42873078 2473
d079837e
SP
2474 if (!index) {
2475 if (open_pack_index(p))
2476 return 0;
2477 level1_ofs = p->index_data;
2478 index = p->index_data;
2479 }
74e34e1f
NP
2480 if (p->index_version > 1) {
2481 level1_ofs += 2;
2482 index += 8;
2483 }
42873078 2484 index += 4 * 256;
74e34e1f
NP
2485 hi = ntohl(level1_ofs[*sha1]);
2486 lo = ((*sha1 == 0x0) ? 0 : ntohl(level1_ofs[*sha1 - 1]));
628522ec
JH
2487 if (p->index_version > 1) {
2488 stride = 20;
2489 } else {
2490 stride = 24;
2491 index += 4;
2492 }
2493
2494 if (debug_lookup)
6e1c2344 2495 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32"\n",
628522ec
JH
2496 sha1[0], sha1[1], sha1[2], lo, hi, p->num_objects);
2497
2498 if (use_lookup < 0)
2499 use_lookup = !!getenv("GIT_USE_LOOKUP");
2500 if (use_lookup) {
2501 int pos = sha1_entry_pos(index, stride, 0,
2502 lo, hi, p->num_objects, sha1);
2503 if (pos < 0)
2504 return 0;
2505 return nth_packed_object_offset(p, pos);
2506 }
1f688557
JH
2507
2508 do {
74e34e1f 2509 unsigned mi = (lo + hi) / 2;
628522ec
JH
2510 int cmp = hashcmp(index + mi * stride, sha1);
2511
2512 if (debug_lookup)
2513 printf("lo %u hi %u rg %u mi %u\n",
2514 lo, hi, hi - lo, mi);
43057304 2515 if (!cmp)
74e34e1f 2516 return nth_packed_object_offset(p, mi);
1f688557
JH
2517 if (cmp > 0)
2518 hi = mi;
2519 else
2520 lo = mi+1;
2521 } while (lo < hi);
2522 return 0;
2523}
2524
4c080182 2525int is_pack_valid(struct packed_git *p)
d131b7af
SP
2526{
2527 /* An already open pack is known to be valid. */
2528 if (p->pack_fd != -1)
2529 return 1;
2530
2531 /* If the pack has one window completely covering the
2532 * file size, the pack is known to be valid even if
2533 * the descriptor is not currently open.
2534 */
2535 if (p->windows) {
2536 struct pack_window *w = p->windows;
2537
2538 if (!w->offset && w->len == p->pack_size)
2539 return 1;
2540 }
2541
2542 /* Force the pack to open to prove its valid. */
2543 return !open_packed_git(p);
2544}
2545
95099731
NTND
2546static int fill_pack_entry(const unsigned char *sha1,
2547 struct pack_entry *e,
2548 struct packed_git *p)
2549{
2550 off_t offset;
2551
2552 if (p->num_bad_objects) {
2553 unsigned i;
2554 for (i = 0; i < p->num_bad_objects; i++)
2555 if (!hashcmp(sha1, p->bad_object_sha1 + 20 * i))
2556 return 0;
2557 }
2558
2559 offset = find_pack_entry_one(sha1, p);
2560 if (!offset)
2561 return 0;
2562
2563 /*
2564 * We are about to tell the caller where they can locate the
2565 * requested object. We better make sure the packfile is
2566 * still here and can be accessed before supplying that
2567 * answer, as it may have been deleted since the index was
2568 * loaded!
2569 */
319b678a 2570 if (!is_pack_valid(p))
95099731 2571 return 0;
95099731
NTND
2572 e->offset = offset;
2573 e->p = p;
2574 hashcpy(e->sha1, sha1);
2575 return 1;
2576}
2577
d40d535b
MH
2578/*
2579 * Iff a pack file contains the object named by sha1, return true and
2580 * store its location to e.
2581 */
4d6acb70 2582static int find_pack_entry(const unsigned char *sha1, struct pack_entry *e)
1f688557
JH
2583{
2584 struct packed_git *p;
43057304 2585
1f688557 2586 prepare_packed_git();
f7c22cc6
NP
2587 if (!packed_git)
2588 return 0;
1f688557 2589
c01f51cc
NTND
2590 if (last_found_pack && fill_pack_entry(sha1, e, last_found_pack))
2591 return 1;
8eca0b47 2592
c01f51cc 2593 for (p = packed_git; p; p = p->next) {
1b1005d1
MH
2594 if (p == last_found_pack)
2595 continue; /* we already checked this one */
f7c22cc6 2596
1b1005d1
MH
2597 if (fill_pack_entry(sha1, e, p)) {
2598 last_found_pack = p;
2599 return 1;
2600 }
c01f51cc 2601 }
1f688557
JH
2602 return 0;
2603}
2604
a6080a0a 2605struct packed_git *find_sha1_pack(const unsigned char *sha1,
bf592c50
DB
2606 struct packed_git *packs)
2607{
2608 struct packed_git *p;
bf592c50
DB
2609
2610 for (p = packs; p; p = p->next) {
43057304 2611 if (find_pack_entry_one(sha1, p))
bf592c50
DB
2612 return p;
2613 }
2614 return NULL;
21666f1a 2615
bf592c50
DB
2616}
2617
052fe5ea 2618static int sha1_loose_object_info(const unsigned char *sha1,
46f03448
KN
2619 struct object_info *oi,
2620 int flags)
65c2e0c3 2621{
46f03448
KN
2622 int status = 0;
2623 unsigned long mapsize;
65c2e0c3 2624 void *map;
ef49a7a0 2625 git_zstream stream;
d65a16f6 2626 char hdr[32];
46f03448 2627 struct strbuf hdrbuf = STRBUF_INIT;
65c2e0c3 2628
5d642e75
JK
2629 if (oi->delta_base_sha1)
2630 hashclr(oi->delta_base_sha1);
2631
052fe5ea
JK
2632 /*
2633 * If we don't care about type or size, then we don't
4ef8d1dd
JH
2634 * need to look inside the object at all. Note that we
2635 * do not optimize out the stat call, even if the
2636 * caller doesn't care about the disk-size, since our
2637 * return value implicitly indicates whether the
2638 * object even exists.
052fe5ea 2639 */
46f03448 2640 if (!oi->typep && !oi->typename && !oi->sizep) {
4ef8d1dd
JH
2641 struct stat st;
2642 if (stat_sha1_file(sha1, &st) < 0)
2643 return -1;
2644 if (oi->disk_sizep)
23c339c0 2645 *oi->disk_sizep = st.st_size;
052fe5ea
JK
2646 return 0;
2647 }
2648
bb6b8e4f 2649 map = map_sha1_file(sha1, &mapsize);
f0df4ed5 2650 if (!map)
dbea72a8 2651 return -1;
23c339c0
JK
2652 if (oi->disk_sizep)
2653 *oi->disk_sizep = mapsize;
46f03448
KN
2654 if ((flags & LOOKUP_UNKNOWN_OBJECT)) {
2655 if (unpack_sha1_header_to_strbuf(&stream, map, mapsize, hdr, sizeof(hdr), &hdrbuf) < 0)
2656 status = error("unable to unpack %s header with --allow-unknown-type",
2657 sha1_to_hex(sha1));
2658 } else if (unpack_sha1_header(&stream, map, mapsize, hdr, sizeof(hdr)) < 0)
36e4d74a
JH
2659 status = error("unable to unpack %s header",
2660 sha1_to_hex(sha1));
46f03448
KN
2661 if (status < 0)
2662 ; /* Do nothing */
2663 else if (hdrbuf.len) {
2664 if ((status = parse_sha1_header_extended(hdrbuf.buf, oi, flags)) < 0)
2665 status = error("unable to parse %s header with --allow-unknown-type",
2666 sha1_to_hex(sha1));
2667 } else if ((status = parse_sha1_header_extended(hdr, oi, flags)) < 0)
36e4d74a 2668 status = error("unable to parse %s header", sha1_to_hex(sha1));
39c68542 2669 git_inflate_end(&stream);
65c2e0c3 2670 munmap(map, mapsize);
46f03448 2671 if (status && oi->typep)
23c339c0 2672 *oi->typep = status;
46f03448 2673 strbuf_release(&hdrbuf);
052fe5ea 2674 return 0;
65c2e0c3
JH
2675}
2676
de7b5d62 2677int sha1_object_info_extended(const unsigned char *sha1, struct object_info *oi, unsigned flags)
f0df4ed5 2678{
c4d9986f 2679 struct cached_object *co;
f0df4ed5 2680 struct pack_entry e;
5b086407 2681 int rtype;
46f03448 2682 enum object_type real_type;
1f7117ef 2683 const unsigned char *real = lookup_replace_object_extended(sha1, flags);
f0df4ed5 2684
1f7117ef 2685 co = find_cached_object(real);
c4d9986f 2686 if (co) {
5b086407
JK
2687 if (oi->typep)
2688 *(oi->typep) = co->type;
9a490590
JH
2689 if (oi->sizep)
2690 *(oi->sizep) = co->size;
161f00e7
JK
2691 if (oi->disk_sizep)
2692 *(oi->disk_sizep) = 0;
5d642e75
JK
2693 if (oi->delta_base_sha1)
2694 hashclr(oi->delta_base_sha1);
46f03448
KN
2695 if (oi->typename)
2696 strbuf_addstr(oi->typename, typename(co->type));
9a490590 2697 oi->whence = OI_CACHED;
5b086407 2698 return 0;
c4d9986f
NTND
2699 }
2700
1f7117ef 2701 if (!find_pack_entry(real, &e)) {
ddd63e64 2702 /* Most likely it's a loose object. */
46f03448 2703 if (!sha1_loose_object_info(real, oi, flags)) {
9a490590 2704 oi->whence = OI_LOOSE;
5b086407 2705 return 0;
9a490590 2706 }
ddd63e64
SG
2707
2708 /* Not a loose object; someone else may have just packed it. */
f0df4ed5 2709 reprepare_packed_git();
1f7117ef 2710 if (!find_pack_entry(real, &e))
052fe5ea 2711 return -1;
f0df4ed5 2712 }
3d77d877 2713
46f03448
KN
2714 /*
2715 * packed_object_info() does not follow the delta chain to
2716 * find out the real type, unless it is given oi->typep.
2717 */
2718 if (oi->typename && !oi->typep)
2719 oi->typep = &real_type;
2720
23c339c0 2721 rtype = packed_object_info(e.p, e.offset, oi);
412916ee 2722 if (rtype < 0) {
1f7117ef 2723 mark_bad_packed_object(e.p, real);
46f03448
KN
2724 if (oi->typep == &real_type)
2725 oi->typep = NULL;
1f7117ef 2726 return sha1_object_info_extended(real, oi, 0);
5266d369
JH
2727 } else if (in_delta_base_cache(e.p, e.offset)) {
2728 oi->whence = OI_DBCACHED;
9a490590
JH
2729 } else {
2730 oi->whence = OI_PACKED;
2731 oi->u.packed.offset = e.offset;
2732 oi->u.packed.pack = e.p;
2733 oi->u.packed.is_delta = (rtype == OBJ_REF_DELTA ||
2734 rtype == OBJ_OFS_DELTA);
3d77d877 2735 }
46f03448
KN
2736 if (oi->typename)
2737 strbuf_addstr(oi->typename, typename(*oi->typep));
2738 if (oi->typep == &real_type)
2739 oi->typep = NULL;
3d77d877 2740
5b086407 2741 return 0;
f0df4ed5
JS
2742}
2743
3fc0dca9 2744/* returns enum object_type or negative */
9a490590
JH
2745int sha1_object_info(const unsigned char *sha1, unsigned long *sizep)
2746{
5b086407 2747 enum object_type type;
d099b717 2748 struct object_info oi = {NULL};
9a490590 2749
5b086407 2750 oi.typep = &type;
9a490590 2751 oi.sizep = sizep;
de7b5d62 2752 if (sha1_object_info_extended(sha1, &oi, LOOKUP_REPLACE_OBJECT) < 0)
5b086407
JK
2753 return -1;
2754 return type;
9a490590
JH
2755}
2756
21666f1a
NP
2757static void *read_packed_sha1(const unsigned char *sha1,
2758 enum object_type *type, unsigned long *size)
1f688557
JH
2759{
2760 struct pack_entry e;
8eca0b47 2761 void *data;
1f688557 2762
cd673c1f 2763 if (!find_pack_entry(sha1, &e))
1f688557 2764 return NULL;
8eca0b47
NP
2765 data = cache_or_unpack_entry(e.p, e.offset, size, type, 1);
2766 if (!data) {
2767 /*
2768 * We're probably in deep shit, but let's try to fetch
2769 * the required object anyway from another pack or loose.
2770 * This should happen only in the presence of a corrupted
2771 * pack, and is better than failing outright.
2772 */
2773 error("failed to read object %s at offset %"PRIuMAX" from %s",
2774 sha1_to_hex(sha1), (uintmax_t)e.offset, e.p->pack_name);
2775 mark_bad_packed_object(e.p, sha1);
ac939109 2776 data = read_object(sha1, type, size);
8eca0b47
NP
2777 }
2778 return data;
1f688557
JH
2779}
2780
21666f1a
NP
2781int pretend_sha1_file(void *buf, unsigned long len, enum object_type type,
2782 unsigned char *sha1)
d66b37bb
JH
2783{
2784 struct cached_object *co;
2785
21666f1a 2786 hash_sha1_file(buf, len, typename(type), sha1);
d66b37bb
JH
2787 if (has_sha1_file(sha1) || find_cached_object(sha1))
2788 return 0;
c7353967 2789 ALLOC_GROW(cached_objects, cached_object_nr + 1, cached_object_alloc);
d66b37bb
JH
2790 co = &cached_objects[cached_object_nr++];
2791 co->size = len;
21666f1a 2792 co->type = type;
efa13f7b
JH
2793 co->buf = xmalloc(len);
2794 memcpy(co->buf, buf, len);
d66b37bb
JH
2795 hashcpy(co->sha1, sha1);
2796 return 0;
2797}
2798
c2c5b270
CC
2799static void *read_object(const unsigned char *sha1, enum object_type *type,
2800 unsigned long *size)
0fcfd160
LT
2801{
2802 unsigned long mapsize;
2803 void *map, *buf;
d66b37bb
JH
2804 struct cached_object *co;
2805
2806 co = find_cached_object(sha1);
2807 if (co) {
21666f1a 2808 *type = co->type;
d66b37bb 2809 *size = co->size;
182af834 2810 return xmemdupz(co->buf, co->size);
d66b37bb 2811 }
0fcfd160 2812
8276c007
PE
2813 buf = read_packed_sha1(sha1, type, size);
2814 if (buf)
2815 return buf;
bb6b8e4f 2816 map = map_sha1_file(sha1, &mapsize);
0fcfd160 2817 if (map) {
7efbff75 2818 buf = unpack_sha1_file(map, mapsize, type, size, sha1);
0fcfd160
LT
2819 munmap(map, mapsize);
2820 return buf;
2821 }
637cdd9d 2822 reprepare_packed_git();
8276c007 2823 return read_packed_sha1(sha1, type, size);
0fcfd160
LT
2824}
2825
b6c4cecc
JH
2826/*
2827 * This function dies on corrupt objects; the callers who want to
2828 * deal with them should arrange to call read_object() and give error
2829 * messages themselves.
2830 */
5bf29b95
JH
2831void *read_sha1_file_extended(const unsigned char *sha1,
2832 enum object_type *type,
2833 unsigned long *size,
2834 unsigned flag)
ac939109 2835{
3ba7a065 2836 void *data;
b6c4cecc 2837 const struct packed_git *p;
bf93eea0 2838 const unsigned char *repl = lookup_replace_object_extended(sha1, flag);
b6c4cecc 2839
3ba7a065
JH
2840 errno = 0;
2841 data = read_object(repl, type, size);
4bbf5a26 2842 if (data)
b6c4cecc 2843 return data;
68095570 2844
25f3af3f 2845 if (errno && errno != ENOENT)
3ba7a065
JH
2846 die_errno("failed to read object %s", sha1_to_hex(sha1));
2847
68095570 2848 /* die if we replaced an object with one that does not exist */
b6c4cecc 2849 if (repl != sha1)
68095570
CC
2850 die("replacement %s not found for %s",
2851 sha1_to_hex(repl), sha1_to_hex(sha1));
2852
b6c4cecc 2853 if (has_loose_object(repl)) {
30d6c6ea
MH
2854 const char *path = sha1_file_name(sha1);
2855
b6c4cecc
JH
2856 die("loose object %s (stored in %s) is corrupt",
2857 sha1_to_hex(repl), path);
e8b15e61 2858 }
68095570 2859
b6c4cecc
JH
2860 if ((p = has_packed_and_bad(repl)) != NULL)
2861 die("packed object %s (stored in %s) is corrupt",
2862 sha1_to_hex(repl), p->pack_name);
f5552aee 2863
b6c4cecc 2864 return NULL;
ac939109
NP
2865}
2866
40469ee9 2867void *read_object_with_reference(const unsigned char *sha1,
21666f1a 2868 const char *required_type_name,
40469ee9
JH
2869 unsigned long *size,
2870 unsigned char *actual_sha1_return)
f4913f91 2871{
21666f1a 2872 enum object_type type, required_type;
f4913f91
JH
2873 void *buffer;
2874 unsigned long isize;
40469ee9 2875 unsigned char actual_sha1[20];
f4913f91 2876
21666f1a 2877 required_type = type_from_string(required_type_name);
e702496e 2878 hashcpy(actual_sha1, sha1);
40469ee9
JH
2879 while (1) {
2880 int ref_length = -1;
2881 const char *ref_type = NULL;
f4913f91 2882
21666f1a 2883 buffer = read_sha1_file(actual_sha1, &type, &isize);
40469ee9
JH
2884 if (!buffer)
2885 return NULL;
21666f1a 2886 if (type == required_type) {
40469ee9
JH
2887 *size = isize;
2888 if (actual_sha1_return)
e702496e 2889 hashcpy(actual_sha1_return, actual_sha1);
40469ee9
JH
2890 return buffer;
2891 }
2892 /* Handle references */
21666f1a 2893 else if (type == OBJ_COMMIT)
40469ee9 2894 ref_type = "tree ";
21666f1a 2895 else if (type == OBJ_TAG)
40469ee9
JH
2896 ref_type = "object ";
2897 else {
2898 free(buffer);
2899 return NULL;
2900 }
2901 ref_length = strlen(ref_type);
f4913f91 2902
50974ec9
MK
2903 if (ref_length + 40 > isize ||
2904 memcmp(buffer, ref_type, ref_length) ||
1d7f171c 2905 get_sha1_hex((char *) buffer + ref_length, actual_sha1)) {
40469ee9
JH
2906 free(buffer);
2907 return NULL;
2908 }
1cf58e72 2909 free(buffer);
40469ee9
JH
2910 /* Now we have the ID of the referred-to object in
2911 * actual_sha1. Check again. */
f4913f91 2912 }
f4913f91
JH
2913}
2914
ce9fbf16 2915static void write_sha1_file_prepare(const void *buf, unsigned long len,
972a9155 2916 const char *type, unsigned char *sha1,
d65a16f6 2917 char *hdr, int *hdrlen)
d410c0f5 2918{
9126f009 2919 git_SHA_CTX c;
d410c0f5
JH
2920
2921 /* Generate the header */
ef1286d3 2922 *hdrlen = xsnprintf(hdr, *hdrlen, "%s %lu", type, len)+1;
d410c0f5
JH
2923
2924 /* Sha1.. */
9126f009
NP
2925 git_SHA1_Init(&c);
2926 git_SHA1_Update(&c, hdr, *hdrlen);
2927 git_SHA1_Update(&c, buf, len);
2928 git_SHA1_Final(sha1, &c);
d410c0f5
JH
2929}
2930
230f1322 2931/*
5a688fe4 2932 * Move the just written object into its final resting place.
230f1322 2933 */
cb5add58 2934int finalize_object_file(const char *tmpfile, const char *filename)
230f1322 2935{
e32c0a9c 2936 int ret = 0;
5a688fe4 2937
348df166 2938 if (object_creation_mode == OBJECT_CREATION_USES_RENAMES)
be66a6c4
JS
2939 goto try_rename;
2940 else if (link(tmpfile, filename))
e32c0a9c 2941 ret = errno;
7ebb6fca
LT
2942
2943 /*
2944 * Coda hack - coda doesn't like cross-directory links,
2945 * so we fall back to a rename, which will mean that it
2946 * won't be able to check collisions, but that's not a
2947 * big deal.
2948 *
2949 * The same holds for FAT formatted media.
2950 *
3be1f18e 2951 * When this succeeds, we just return. We have nothing
7ebb6fca
LT
2952 * left to unlink.
2953 */
2954 if (ret && ret != EEXIST) {
be66a6c4 2955 try_rename:
7ebb6fca 2956 if (!rename(tmpfile, filename))
3be1f18e 2957 goto out;
9e48b389 2958 ret = errno;
230f1322 2959 }
691f1a28 2960 unlink_or_warn(tmpfile);
230f1322
LT
2961 if (ret) {
2962 if (ret != EEXIST) {
82247e9b 2963 return error("unable to write sha1 filename %s: %s", filename, strerror(ret));
230f1322
LT
2964 }
2965 /* FIXME!!! Collision check here ? */
2966 }
2967
3be1f18e 2968out:
5256b006 2969 if (adjust_shared_perm(filename))
5a688fe4 2970 return error("unable to set permission to '%s'", filename);
230f1322
LT
2971 return 0;
2972}
2973
4d548150
LT
2974static int write_buffer(int fd, const void *buf, size_t len)
2975{
d34cf19b 2976 if (write_in_full(fd, buf, len) < 0)
93822c22 2977 return error("file write error (%s)", strerror(errno));
4d548150
LT
2978 return 0;
2979}
2980
ce9fbf16 2981int hash_sha1_file(const void *buf, unsigned long len, const char *type,
abdc3fc8
RS
2982 unsigned char *sha1)
2983{
d65a16f6 2984 char hdr[32];
ef1286d3 2985 int hdrlen = sizeof(hdr);
abdc3fc8
RS
2986 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
2987 return 0;
2988}
2989
e9039dd3
LT
2990/* Finalize a file on disk, and close it. */
2991static void close_sha1_file(int fd)
2992{
aafe9fba
LT
2993 if (fsync_object_files)
2994 fsync_or_die(fd, "sha1 file");
e9039dd3 2995 if (close(fd) != 0)
d824cbba 2996 die_errno("error when closing sha1 file");
e9039dd3
LT
2997}
2998
5723fe7e
LT
2999/* Size of directory component, including the ending '/' */
3000static inline int directory_size(const char *filename)
3001{
3002 const char *s = strrchr(filename, '/');
3003 if (!s)
3004 return 0;
3005 return s - filename + 1;
3006}
3007
3008/*
3009 * This creates a temporary file in the same directory as the final
3010 * 'filename'
3011 *
3012 * We want to avoid cross-directory filename renames, because those
3013 * can have problems on various filesystems (FAT, NFS, Coda).
3014 */
d4b3d11a 3015static int create_tmpfile(struct strbuf *tmp, const char *filename)
5723fe7e
LT
3016{
3017 int fd, dirlen = directory_size(filename);
3018
d4b3d11a
JK
3019 strbuf_reset(tmp);
3020 strbuf_add(tmp, filename, dirlen);
3021 strbuf_addstr(tmp, "tmp_obj_XXXXXX");
3022 fd = git_mkstemp_mode(tmp->buf, 0444);
cbacbf4e 3023 if (fd < 0 && dirlen && errno == ENOENT) {
d4b3d11a
JK
3024 /*
3025 * Make sure the directory exists; note that the contents
3026 * of the buffer are undefined after mkstemp returns an
3027 * error, so we have to rewrite the whole buffer from
3028 * scratch.
3029 */
3030 strbuf_reset(tmp);
3031 strbuf_add(tmp, filename, dirlen - 1);
3032 if (mkdir(tmp->buf, 0777) && errno != EEXIST)
b2476a60 3033 return -1;
d4b3d11a 3034 if (adjust_shared_perm(tmp->buf))
5723fe7e
LT
3035 return -1;
3036
3037 /* Try again */
d4b3d11a
JK
3038 strbuf_addstr(tmp, "/tmp_obj_XXXXXX");
3039 fd = git_mkstemp_mode(tmp->buf, 0444);
5723fe7e
LT
3040 }
3041 return fd;
3042}
3043
bbac7311 3044static int write_loose_object(const unsigned char *sha1, char *hdr, int hdrlen,
c00e657d 3045 const void *buf, unsigned long len, time_t mtime)
0fcfd160 3046{
915308b1 3047 int fd, ret;
9892beba 3048 unsigned char compressed[4096];
ef49a7a0 3049 git_zstream stream;
748af44c
NP
3050 git_SHA_CTX c;
3051 unsigned char parano_sha1[20];
d4b3d11a 3052 static struct strbuf tmp_file = STRBUF_INIT;
30d6c6ea 3053 const char *filename = sha1_file_name(sha1);
a44c9a5e 3054
d4b3d11a 3055 fd = create_tmpfile(&tmp_file, filename);
aac17941 3056 if (fd < 0) {
35243577 3057 if (errno == EACCES)
82247e9b 3058 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
916d081b 3059 else
5eaeda70 3060 return error("unable to create temporary file: %s", strerror(errno));
aac17941
LT
3061 }
3062
0fcfd160 3063 /* Set it up */
55bb5c91 3064 git_deflate_init(&stream, zlib_compression_level);
0fcfd160 3065 stream.next_out = compressed;
9892beba 3066 stream.avail_out = sizeof(compressed);
748af44c 3067 git_SHA1_Init(&c);
a44c9a5e
LT
3068
3069 /* First header.. */
d65a16f6 3070 stream.next_in = (unsigned char *)hdr;
a44c9a5e 3071 stream.avail_in = hdrlen;
55bb5c91
JH
3072 while (git_deflate(&stream, 0) == Z_OK)
3073 ; /* nothing */
748af44c 3074 git_SHA1_Update(&c, hdr, hdrlen);
a44c9a5e
LT
3075
3076 /* Then the data itself.. */
c00e657d 3077 stream.next_in = (void *)buf;
a44c9a5e 3078 stream.avail_in = len;
9892beba 3079 do {
748af44c 3080 unsigned char *in0 = stream.next_in;
55bb5c91 3081 ret = git_deflate(&stream, Z_FINISH);
748af44c 3082 git_SHA1_Update(&c, in0, stream.next_in - in0);
9892beba
NP
3083 if (write_buffer(fd, compressed, stream.next_out - compressed) < 0)
3084 die("unable to write sha1 file");
3085 stream.next_out = compressed;
3086 stream.avail_out = sizeof(compressed);
3087 } while (ret == Z_OK);
3088
ac54c277
LT
3089 if (ret != Z_STREAM_END)
3090 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1), ret);
55bb5c91 3091 ret = git_deflate_end_gently(&stream);
ac54c277
LT
3092 if (ret != Z_OK)
3093 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1), ret);
748af44c
NP
3094 git_SHA1_Final(parano_sha1, &c);
3095 if (hashcmp(sha1, parano_sha1) != 0)
3096 die("confused by unstable object source data for %s", sha1_to_hex(sha1));
ac54c277 3097
e9039dd3 3098 close_sha1_file(fd);
0fcfd160 3099
bbac7311
NP
3100 if (mtime) {
3101 struct utimbuf utb;
3102 utb.actime = mtime;
3103 utb.modtime = mtime;
d4b3d11a 3104 if (utime(tmp_file.buf, &utb) < 0)
bbac7311 3105 warning("failed utime() on %s: %s",
d4b3d11a 3106 tmp_file.buf, strerror(errno));
bbac7311
NP
3107 }
3108
d4b3d11a 3109 return finalize_object_file(tmp_file.buf, filename);
0fcfd160 3110}
8237b185 3111
33d4221c
JK
3112static int freshen_loose_object(const unsigned char *sha1)
3113{
3114 return check_and_freshen(sha1, 1);
3115}
3116
3117static int freshen_packed_object(const unsigned char *sha1)
3118{
3119 struct pack_entry e;
ee1c6c34
JK
3120 if (!find_pack_entry(sha1, &e))
3121 return 0;
3122 if (e.p->freshened)
3123 return 1;
3124 if (!freshen_file(e.p->pack_name))
3125 return 0;
3126 e.p->freshened = 1;
3127 return 1;
33d4221c
JK
3128}
3129
1427a7ff 3130int write_sha1_file(const void *buf, unsigned long len, const char *type, unsigned char *sha1)
bbac7311 3131{
bbac7311 3132 char hdr[32];
ef1286d3 3133 int hdrlen = sizeof(hdr);
bbac7311
NP
3134
3135 /* Normally if we have it in the pack then we do not bother writing
3136 * it out into .git/objects/??/?{38} file.
3137 */
3138 write_sha1_file_prepare(buf, len, type, sha1, hdr, &hdrlen);
b5f52f37 3139 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
bbac7311
NP
3140 return 0;
3141 return write_loose_object(sha1, hdr, hdrlen, buf, len, 0);
3142}
3143
0c3db67c
ES
3144int hash_sha1_file_literally(const void *buf, unsigned long len, const char *type,
3145 unsigned char *sha1, unsigned flags)
3146{
3147 char *header;
3148 int hdrlen, status = 0;
3149
3150 /* type string, SP, %lu of the length plus NUL must fit this */
ef1286d3
JK
3151 hdrlen = strlen(type) + 32;
3152 header = xmalloc(hdrlen);
0c3db67c
ES
3153 write_sha1_file_prepare(buf, len, type, sha1, header, &hdrlen);
3154
3155 if (!(flags & HASH_WRITE_OBJECT))
3156 goto cleanup;
051086b9 3157 if (freshen_packed_object(sha1) || freshen_loose_object(sha1))
0c3db67c
ES
3158 goto cleanup;
3159 status = write_loose_object(sha1, header, hdrlen, buf, len, 0);
3160
3161cleanup:
3162 free(header);
3163 return status;
3164}
3165
bbac7311
NP
3166int force_object_loose(const unsigned char *sha1, time_t mtime)
3167{
bbac7311
NP
3168 void *buf;
3169 unsigned long len;
3170 enum object_type type;
3171 char hdr[32];
3172 int hdrlen;
1fb23e65 3173 int ret;
bbac7311 3174
c529d75a 3175 if (has_loose_object(sha1))
bbac7311
NP
3176 return 0;
3177 buf = read_packed_sha1(sha1, &type, &len);
3178 if (!buf)
3179 return error("cannot read sha1_file for %s", sha1_to_hex(sha1));
ef1286d3 3180 hdrlen = xsnprintf(hdr, sizeof(hdr), "%s %lu", typename(type), len) + 1;
1fb23e65
BS
3181 ret = write_loose_object(sha1, hdr, hdrlen, buf, len, mtime);
3182 free(buf);
3183
3184 return ret;
bbac7311
NP
3185}
3186
bf592c50
DB
3187int has_pack_index(const unsigned char *sha1)
3188{
3189 struct stat st;
3190 if (stat(sha1_pack_index_name(sha1), &st))
3191 return 0;
3192 return 1;
bf592c50
DB
3193}
3194
cd673c1f
JH
3195int has_sha1_pack(const unsigned char *sha1)
3196{
3197 struct pack_entry e;
3198 return find_pack_entry(sha1, &e);
3199}
3200
0eeb077b 3201int has_sha1_file_with_flags(const unsigned char *sha1, int flags)
8237b185 3202{
1f688557
JH
3203 struct pack_entry e;
3204
cd673c1f 3205 if (find_pack_entry(sha1, &e))
1f688557 3206 return 1;
45e8a748
JK
3207 if (has_loose_object(sha1))
3208 return 1;
0eeb077b
JK
3209 if (flags & HAS_SHA1_QUICK)
3210 return 0;
45e8a748
JK
3211 reprepare_packed_git();
3212 return find_pack_entry(sha1, &e);
8237b185 3213}
74400e71 3214
c879daa2
NTND
3215static void check_tree(const void *buf, size_t size)
3216{
3217 struct tree_desc desc;
3218 struct name_entry entry;
3219
3220 init_tree_desc(&desc, buf, size);
3221 while (tree_entry(&desc, &entry))
3222 /* do nothing
3223 * tree_entry() will die() on malformed entries */
3224 ;
3225}
3226
3227static void check_commit(const void *buf, size_t size)
3228{
3229 struct commit c;
3230 memset(&c, 0, sizeof(c));
3231 if (parse_commit_buffer(&c, buf, size))
3232 die("corrupt commit");
3233}
3234
3235static void check_tag(const void *buf, size_t size)
3236{
3237 struct tag t;
3238 memset(&t, 0, sizeof(t));
3239 if (parse_tag_buffer(&t, buf, size))
3240 die("corrupt tag");
3241}
3242
43df4f86 3243static int index_mem(unsigned char *sha1, void *buf, size_t size,
c4ce46fc
JH
3244 enum object_type type,
3245 const char *path, unsigned flags)
e7332f96 3246{
6c510bee 3247 int ret, re_allocated = 0;
c4ce46fc 3248 int write_object = flags & HASH_WRITE_OBJECT;
74400e71 3249
7672db20 3250 if (!type)
edaec3fb 3251 type = OBJ_BLOB;
6c510bee
LT
3252
3253 /*
3254 * Convert blobs to git internal format
3255 */
43df4f86 3256 if ((type == OBJ_BLOB) && path) {
f285a2d7 3257 struct strbuf nbuf = STRBUF_INIT;
21e5ad50 3258 if (convert_to_git(path, buf, size, &nbuf,
5e12e78e 3259 write_object ? safe_crlf : SAFE_CRLF_FALSE)) {
b315c5c0 3260 buf = strbuf_detach(&nbuf, &size);
6c510bee
LT
3261 re_allocated = 1;
3262 }
3263 }
c4ce46fc 3264 if (flags & HASH_FORMAT_CHECK) {
c879daa2
NTND
3265 if (type == OBJ_TREE)
3266 check_tree(buf, size);
3267 if (type == OBJ_COMMIT)
3268 check_commit(buf, size);
3269 if (type == OBJ_TAG)
3270 check_tag(buf, size);
3271 }
6c510bee 3272
7672db20 3273 if (write_object)
edaec3fb 3274 ret = write_sha1_file(buf, size, typename(type), sha1);
abdc3fc8 3275 else
edaec3fb 3276 ret = hash_sha1_file(buf, size, typename(type), sha1);
43df4f86 3277 if (re_allocated)
6c510bee 3278 free(buf);
43df4f86
DP
3279 return ret;
3280}
3281
9035d75a
SP
3282static int index_stream_convert_blob(unsigned char *sha1, int fd,
3283 const char *path, unsigned flags)
3284{
3285 int ret;
3286 const int write_object = flags & HASH_WRITE_OBJECT;
3287 struct strbuf sbuf = STRBUF_INIT;
3288
3289 assert(path);
3290 assert(would_convert_to_git_filter_fd(path));
3291
3292 convert_to_git_filter_fd(path, fd, &sbuf,
3293 write_object ? safe_crlf : SAFE_CRLF_FALSE);
3294
3295 if (write_object)
3296 ret = write_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3297 sha1);
3298 else
3299 ret = hash_sha1_file(sbuf.buf, sbuf.len, typename(OBJ_BLOB),
3300 sha1);
3301 strbuf_release(&sbuf);
3302 return ret;
3303}
3304
7b41e1e1
JH
3305static int index_pipe(unsigned char *sha1, int fd, enum object_type type,
3306 const char *path, unsigned flags)
3307{
3308 struct strbuf sbuf = STRBUF_INIT;
3309 int ret;
3310
3311 if (strbuf_read(&sbuf, fd, 4096) >= 0)
3312 ret = index_mem(sha1, sbuf.buf, sbuf.len, type, path, flags);
3313 else
3314 ret = -1;
3315 strbuf_release(&sbuf);
3316 return ret;
3317}
3318
ea68b0ce
DP
3319#define SMALL_FILE_SIZE (32*1024)
3320
7b41e1e1
JH
3321static int index_core(unsigned char *sha1, int fd, size_t size,
3322 enum object_type type, const char *path,
3323 unsigned flags)
43df4f86
DP
3324{
3325 int ret;
43df4f86 3326
7b41e1e1 3327 if (!size) {
f6a1e1e2 3328 ret = index_mem(sha1, "", size, type, path, flags);
ea68b0ce
DP
3329 } else if (size <= SMALL_FILE_SIZE) {
3330 char *buf = xmalloc(size);
3331 if (size == read_in_full(fd, buf, size))
c4ce46fc 3332 ret = index_mem(sha1, buf, size, type, path, flags);
ea68b0ce
DP
3333 else
3334 ret = error("short read %s", strerror(errno));
3335 free(buf);
08bda208 3336 } else {
43df4f86 3337 void *buf = xmmap(NULL, size, PROT_READ, MAP_PRIVATE, fd, 0);
c4ce46fc 3338 ret = index_mem(sha1, buf, size, type, path, flags);
aac17941 3339 munmap(buf, size);
08bda208 3340 }
7b41e1e1
JH
3341 return ret;
3342}
3343
4dd1fbc7 3344/*
568508e7
JH
3345 * This creates one packfile per large blob unless bulk-checkin
3346 * machinery is "plugged".
4dd1fbc7
JH
3347 *
3348 * This also bypasses the usual "convert-to-git" dance, and that is on
3349 * purpose. We could write a streaming version of the converting
3350 * functions and insert that before feeding the data to fast-import
4f22b101
JK
3351 * (or equivalent in-core API described above). However, that is
3352 * somewhat complicated, as we do not know the size of the filter
3353 * result, which we need to know beforehand when writing a git object.
3354 * Since the primary motivation for trying to stream from the working
3355 * tree file and to avoid mmaping it in core is to deal with large
3356 * binary blobs, they generally do not want to get any conversion, and
3357 * callers should avoid this code path when filters are requested.
4dd1fbc7
JH
3358 */
3359static int index_stream(unsigned char *sha1, int fd, size_t size,
3360 enum object_type type, const char *path,
3361 unsigned flags)
3362{
568508e7 3363 return index_bulk_checkin(sha1, fd, size, type, path, flags);
4dd1fbc7
JH
3364}
3365
7b41e1e1
JH
3366int index_fd(unsigned char *sha1, int fd, struct stat *st,
3367 enum object_type type, const char *path, unsigned flags)
3368{
3369 int ret;
7b41e1e1 3370
9079ab7c
SP
3371 /*
3372 * Call xsize_t() only when needed to avoid potentially unnecessary
3373 * die() for large files.
3374 */
9035d75a
SP
3375 if (type == OBJ_BLOB && path && would_convert_to_git_filter_fd(path))
3376 ret = index_stream_convert_blob(sha1, fd, path, flags);
3377 else if (!S_ISREG(st->st_mode))
7b41e1e1 3378 ret = index_pipe(sha1, fd, type, path, flags);
9079ab7c 3379 else if (st->st_size <= big_file_threshold || type != OBJ_BLOB ||
7ce7c760 3380 (path && would_convert_to_git(path)))
9079ab7c
SP
3381 ret = index_core(sha1, fd, xsize_t(st->st_size), type, path,
3382 flags);
4dd1fbc7 3383 else
9079ab7c
SP
3384 ret = index_stream(sha1, fd, xsize_t(st->st_size), type, path,
3385 flags);
43df4f86 3386 close(fd);
aac17941 3387 return ret;
74400e71 3388}
ec1fcc16 3389
c4ce46fc 3390int index_path(unsigned char *sha1, const char *path, struct stat *st, unsigned flags)
ec1fcc16
JH
3391{
3392 int fd;
b760d3aa 3393 struct strbuf sb = STRBUF_INIT;
ec1fcc16
JH
3394
3395 switch (st->st_mode & S_IFMT) {
3396 case S_IFREG:
3397 fd = open(path, O_RDONLY);
3398 if (fd < 0)
3399 return error("open(\"%s\"): %s", path,
3400 strerror(errno));
c4ce46fc 3401 if (index_fd(sha1, fd, st, OBJ_BLOB, path, flags) < 0)
ec1fcc16
JH
3402 return error("%s: failed to insert into database",
3403 path);
3404 break;
3405 case S_IFLNK:
b760d3aa 3406 if (strbuf_readlink(&sb, path, st->st_size)) {
ec1fcc16 3407 char *errstr = strerror(errno);
ec1fcc16
JH
3408 return error("readlink(\"%s\"): %s", path,
3409 errstr);
3410 }
c4ce46fc 3411 if (!(flags & HASH_WRITE_OBJECT))
b760d3aa
LT
3412 hash_sha1_file(sb.buf, sb.len, blob_type, sha1);
3413 else if (write_sha1_file(sb.buf, sb.len, blob_type, sha1))
ec1fcc16
JH
3414 return error("%s: failed to insert into database",
3415 path);
b760d3aa 3416 strbuf_release(&sb);
ec1fcc16 3417 break;
f35a6d3b
LT
3418 case S_IFDIR:
3419 return resolve_gitlink_ref(path, "HEAD", sha1);
ec1fcc16
JH
3420 default:
3421 return error("%s: unsupported file type", path);
3422 }
3423 return 0;
3424}
a69e5429
JH
3425
3426int read_pack_header(int fd, struct pack_header *header)
3427{
c697ad14
HO
3428 if (read_in_full(fd, header, sizeof(*header)) < sizeof(*header))
3429 /* "eof before pack header was fully read" */
3430 return PH_ERROR_EOF;
3431
a69e5429
JH
3432 if (header->hdr_signature != htonl(PACK_SIGNATURE))
3433 /* "protocol error (pack signature mismatch detected)" */
3434 return PH_ERROR_PACK_SIGNATURE;
3435 if (!pack_version_ok(header->hdr_version))
3436 /* "protocol error (pack version unsupported)" */
3437 return PH_ERROR_PROTOCOL;
3438 return 0;
3439}
40d52ff7
JK
3440
3441void assert_sha1_type(const unsigned char *sha1, enum object_type expect)
3442{
3443 enum object_type type = sha1_object_info(sha1, NULL);
3444 if (type < 0)
3445 die("%s is not a valid object", sha1_to_hex(sha1));
3446 if (type != expect)
3447 die("%s is not a valid '%s' object", sha1_to_hex(sha1),
3448 typename(expect));
3449}
27e1e22d
JK
3450
3451static int for_each_file_in_obj_subdir(int subdir_nr,
3452 struct strbuf *path,
3453 each_loose_object_fn obj_cb,
3454 each_loose_cruft_fn cruft_cb,
3455 each_loose_subdir_fn subdir_cb,
3456 void *data)
3457{
3458 size_t baselen = path->len;
3459 DIR *dir = opendir(path->buf);
3460 struct dirent *de;
3461 int r = 0;
3462
3463 if (!dir) {
3464 if (errno == ENOENT)
3465 return 0;
3466 return error("unable to open %s: %s", path->buf, strerror(errno));
3467 }
3468
3469 while ((de = readdir(dir))) {
3470 if (is_dot_or_dotdot(de->d_name))
3471 continue;
3472
3473 strbuf_setlen(path, baselen);
3474 strbuf_addf(path, "/%s", de->d_name);
3475
3476 if (strlen(de->d_name) == 38) {
3477 char hex[41];
3478 unsigned char sha1[20];
3479
3480 snprintf(hex, sizeof(hex), "%02x%s",
3481 subdir_nr, de->d_name);
3482 if (!get_sha1_hex(hex, sha1)) {
3483 if (obj_cb) {
3484 r = obj_cb(sha1, path->buf, data);
3485 if (r)
3486 break;
3487 }
3488 continue;
3489 }
3490 }
3491
3492 if (cruft_cb) {
3493 r = cruft_cb(de->d_name, path->buf, data);
3494 if (r)
3495 break;
3496 }
3497 }
3498 strbuf_setlen(path, baselen);
3499
3500 if (!r && subdir_cb)
3501 r = subdir_cb(subdir_nr, path->buf, data);
3502
3503 closedir(dir);
3504 return r;
3505}
3506
e6f875e0 3507int for_each_loose_file_in_objdir_buf(struct strbuf *path,
27e1e22d
JK
3508 each_loose_object_fn obj_cb,
3509 each_loose_cruft_fn cruft_cb,
3510 each_loose_subdir_fn subdir_cb,
3511 void *data)
3512{
e6f875e0 3513 size_t baselen = path->len;
27e1e22d
JK
3514 int r = 0;
3515 int i;
3516
27e1e22d 3517 for (i = 0; i < 256; i++) {
e6f875e0
JK
3518 strbuf_addf(path, "/%02x", i);
3519 r = for_each_file_in_obj_subdir(i, path, obj_cb, cruft_cb,
27e1e22d 3520 subdir_cb, data);
e6f875e0 3521 strbuf_setlen(path, baselen);
27e1e22d
JK
3522 if (r)
3523 break;
3524 }
3525
e6f875e0
JK
3526 return r;
3527}
3528
3529int for_each_loose_file_in_objdir(const char *path,
3530 each_loose_object_fn obj_cb,
3531 each_loose_cruft_fn cruft_cb,
3532 each_loose_subdir_fn subdir_cb,
3533 void *data)
3534{
3535 struct strbuf buf = STRBUF_INIT;
3536 int r;
3537
3538 strbuf_addstr(&buf, path);
3539 r = for_each_loose_file_in_objdir_buf(&buf, obj_cb, cruft_cb,
3540 subdir_cb, data);
27e1e22d 3541 strbuf_release(&buf);
e6f875e0 3542
27e1e22d
JK
3543 return r;
3544}
660c889e
JK
3545
3546struct loose_alt_odb_data {
3547 each_loose_object_fn *cb;
3548 void *data;
3549};
3550
3551static int loose_from_alt_odb(struct alternate_object_database *alt,
3552 void *vdata)
3553{
3554 struct loose_alt_odb_data *data = vdata;
b0a42642
JM
3555 struct strbuf buf = STRBUF_INIT;
3556 int r;
3557
3558 /* copy base not including trailing '/' */
3559 strbuf_add(&buf, alt->base, alt->name - alt->base - 1);
3560